Linus Thinks Virtualization Is 'Evil'

Front page first-timer crdotson writes "Linus said in an interview that he thinks virtualization is 'evil' because he prefers to deal with the real hardware. Hardware virtualization allows for better barriers between systems by running multiple OSes on the same hardware, but OS-level virtualization allows similar barriers without a hypervisor between the kernel and the hardware. Should we expect more focus on OS-level virtualization such as Linux-VServer, OpenVZ, and LXC?"

Comments

Some might argue

[Anonymous Coward]

That your OS being tied to a particular piece of hardware without a ton of effort is also "evil." Migration is one of the best things ever.

Re:Some might argue

[h4rr4r]

Sounds like your OS needs fixed. I have migrated linux boxes with dd.

Re:Some might argue

[alphatel]

Universe: Replace hardware, rebuild server + much grunting
Multiverse: Replace Hardware, go to sleep.

Re:Some might argue

[FaxeTheCat]

While they are running?

Re:Some might argue

[MightyMartian]

As have I, and ended up going through a real pain in the ass with udev. Not as bad as migrating Windows, of course, at least Linux almost always boots (in the old days when I used Slackware I used to compile a braindead generic kernel that would pretty much boot on anything and add it to the LILO menu), but still, depending on the hardware, it can be a hastle.

With virtualization, as long as I stick to a particular virtualization system (generally KVM), so far as the guest OS is concerned, it doesn't know the difference. It sees the same virtualized hardware and life is good. That's what I want. I want to be able to move OSs, regardless of how easy or difficult it may be to migrate any particular OS, without the hastle.

Re:Some might argue

[b0bby]

Sounds like your OS needs fixed. I have migrated linux boxes with dd.

But did you do that while there were live users on the system?

Re:Some might argue

[Anonymous Coward]

You don't want to dd every system on a box just because you have to reboot. We just had a hardware error that came up on one of our ESX hosts and we needed to reboot to clear it. It took just one click to enter Maintenance Mode and then another to power down the host. All of the guests migrated without any other intervention. Powered the system back online, exited Maintenance Mode, and all the guests flocked back to the host. Very simple and elegant and no users were impacted when the host was restarted.

Re:Some might argue

[Anonymous Coward]

Yep.

Re:Some might argue

[kwark]

If it's running just use tar/rsync.

Re:Some might argue

[turbidostato]

"Sounds like your OS needs fixed. I have migrated linux boxes with dd."

In 2 to 10 seconds? Live? Without people even losing their current open sessions?

I really don't think so.

Re:Some might argue

[vux984]

I think you missed the point.

You can migrate a server in virtualization to new host hardware without shutting it down.

This is not the same as "copy, shutdown the old, bring up the new". This is live migration without a shutdown.

Re:Some might argue

[micsaund]

He was 99.99% likely referring to migrating a live VM to another host, not imaging your desktop to another hard drive. In the enterprise, downtime is often not well tolerated (not something I agree 100% but whatever) and live migration (aka VMware VMotion for example) enables you to take a virtualization host and vacate all of the *still running* VMs off to another host when you need to do maintenance or whatever else to the physical hardware. In the case of VMotion, there is a loss of about one ping on the network while the VM execution is cut-over to the new host, but otherwise, all network connections/etc. remain persistent.

It's very handy stuff, but not something someone who's only familiar with home user or individually installed boxes may be familiar with. It's definitely not just using 'dd'.

Re:Some might argue

[Anonymous Coward]

I think he means that he used dd to simply copy a disk to another computer and it worked without drivers or authentication from microsoft......

You're talking about live migration....He reeks of a 'kiddie and doesn't understand you at all.

Re:Some might argue

[h4rr4r]

Actually I run a vmware setup at work.

Re:Some might argue

[bigredradio]

The biggest problem with that is trying to migrate to dissimilar hardware. Changes to the storage adapter or disk size (smaller) make for real problems.

<shameless plug>Storix is good for migrating to different hardware.</shameless plug>

Re:Some might argue

[h4rr4r]

A kiddie, have not been that in probably longer than this AC has been alive. To me migration means moving to other hardware permanently. Vmotion is vmotion. I use that all the time. Still does not mean that local hardware and dd don't have their place.

Re:Some might argue

[teftin]

Which smells badly designed service. It being running should be tied to any specific hardware/vm/os running.

Re:Some might argue

[Anonymous Coward]

well if you run a vmware setup at work and migrate boxes with dd, i'd like to talk to your manager.

Re:Some might argue

[h4rr4r]

Not those boxes you tard.

Somethings like stuff that is IO limited belong on real hardware.

Re:Some might argue

[Thing 1]

Hi, "hassle" rhymes with "castle" but is spelled differently. Your helpful grammar Ally.

Re:Some might argue

[sjames]

Application level migration would be even cooler. Why should migratable resources be arbitrarily glued together into "system images"?

Re:Some might argue

[vux984]

Which smells badly designed service. It being running should be tied to any specific hardware/vm/os running.

As yes...the old all software ever written is just badly designed argument. We just need to fix all software ever written and then we won't need OS virtualization.

Re:Some might argue

[Anonymous Coward]

Wise man say: "when you find yourself in hole, stop digging."

Re:Some might argue

[cowboy76Spain]

Do not forget to fix customers so they use the services at regular intervals, so there are no more peak hours or days.

I suggest trying with a hammer (it does not work very well, but is very satisfactory).

Re:Some might argue

[h4rr4r]

Stupid AC assume migration always means live migration.

Re:Some might argue

[cowboy76Spain]

Very informative, thank you (HINT: I was raised by a pair of untrained monkeys) :-)

Re:Some might argue

[Lunix Nutcase]

Except that's what it did mean. Keep digging that hole though.

Re:Some might argue

[Anonymous Coward]

This is not the same as "copy, shutdown the old, bring up the new". This is live migration without a shutdown.

Hmmrm, I used to be able to do this with HA-linux by pulling out the network and heartbeat wires on the live box.

But, like virtualization, it was a very complex solution that wasn't usually worth the effort. I ended up making the services clustered and heavily redundant and got rid of HA-linux's complexity.

Well written apps really don't need virtualization. That being said, the modern easy-to-use virtualization systems like vmware are dead awesome, and can make up for some serious architectural problems that other idiots have stuck you with.

Re:Some might argue

[Aighearach]

# wall server migration reconnect in 5 minutes
# shutdown -h +5

<3 Virtualization

Re:Some might argue

[19thNervousBreakdown]

He's only wasting his time if it doesn't work.

Re:Some might argue

[pjt33]

Only if you have a non-rhotic accent.

Re:Some might argue

[hairyfeet]

And I'd argue that Linus is a selfish arrogant asshole that shouldn't be listened to. Look at the facts: In 1993 Linus decides that hardware ABIs are evil and he won't allow them. now at the time the PC was at 30Mhz and RAM was worth more per Mb than diamonds so this view may or may not have been justified depending on the overhead.

But now here it is 2011, and ALL of the major OSes, MSFT Windows, Apple OSX, BSD all flavors, hell even OS/2 has an ABI. This allows a person or company to "write once, use for years" thus making it easier on BOTH the company AND the user who doesn't have to deal with constant driver breakage. Also PCs are multicores with even old machines having 512Mb of RAM, so overhead for an ABI would be trivial. Does Linus admit times have changed and change his tune? Fuck no! He continues to Goatse the kernel anytime he pleases and acts like it is still 1993 and he can do whatever he wants, fuck everybody else.

Crap like this and TFA just shows that Linus is completely out of touch and frankly needs to "pursue other interests" so that Linux can go forward. I can tell you both in my shop and talking to other shops it is the fucking driver mess that keeps anyone from offering Linux because frankly the support costs of having one or more drivers break every 6 months would drive our costs through the roof. I can also give links to articles on big name corps like Walamrt and ASUS bailing on Linux. Why? same thing fiddly driver bullshit shoots up support costs.

So can we PLEASE have someone fork the kernel already? that is what is supposed to happen in FOSS when a sitch gets bad, fork right? Well I'd argue the sitch with Linus is bad with a capital B. he was great back in the day but we ain't in back in the day anymore. He doesn't get VMs and he doesn't get ABIs, what else doesn't he get? I'd argue if it weren't for driver borkage Linux could be making serious inroads. there are plenty of whitebox shops that would be happy to get rid of Windows licensing costs, plenty of places like Walmart that are always looking for a way to lower the price. No Windows tax? means I can undercut my competition by a cool $100 right off the top.

But until Linux gets someone at the helm that doesn't treat the kernel as his personal playtoy and thinks about the consumers Linux will stay stuck at 1%.

Linux is a good OS, it has great DEs, tons of nice software, good security, frankly it deserves better than Torvalds and his kernel screwing.

Re:Some might argue

[cloudmaster]

I just pulled the hard drive out of a computer with a Pentium II, carried it to a different computer with all different hardware (including a P3 processor), and turned it on. Everything works exactly as before. Hooray for the adequate level of abstraction a properly configured udev provides!

Re:Some might argue

[cloudmaster]

You should start using UUIDs or disk labels in fstab, and should consider building modular kernels. Maybe check in to a SAN architecture as well. :)

Re:Some might argue

[MightyMartian]

And that's marvelous for a desktop machine. But I've built enough custom routers using Linux to know that udev can be a right pain in the ass, because where you have multiple NICs, all of a sudden you have things being rejigged as eth2 or whatever. Certainly fixable, but still a pain.

Re:Some might argue

[Bucky24]

Maybe I'm crazy, but I actually really enjoy rebuilding servers (software wise, not hardware). At my workplace we are slowly moving to virtual slices instead of actual servers. Takes away all the fun of installing the OS and configuring everything, IMO. A lot quicker though... I guess that's the point.

Re:Some might argue

[crutchy]

datacentres use virtualization merely to squeeze more servers out of their existing hardware (equals more paying customers) so virtualization for anything else is really just a fad, and it has nothing to do with qos or uptime (which is more about hardware failover and eliminating spof - unless you use windows in which case the operating system is the single point of failure). customers care about "service", not "servers". redirecting a service to different hardware doesn't require virtualisation; more experienced geeks would no doubt have a much more elegant solution (that could be performed via ssh) but the simplest approach would be to disconnect the network cable from the old server and plug it into a ready-configured new replacement. i'm aware of tools like heartbeat and pacemaker for automated/remote failover, but i'm not experienced with that.

Re:Some might argue

[grcumb]

Hi, "hassle" rhymes with "castle" but is spelled differently. Your helpful grammar Ally.

Nono, you're mistaken. It's from an old Hippie saying:

A man's comb is his hastle.

Re:Some might argue

[bigredradio]

Did you mean monolithic kernels? The fact that there are different modules to load IS the problem. The initrd needs to get access to the root filesystem and if the correct controller module is not loaded....no root. Same problem with SAN. If you change HBA cards you may need different modules to load. But most likely the SAN handler will be the same.

Re:Some might argue

[crutchy]

Re: "In 2 to 10 seconds? Live? Without people even losing their current open sessions?" i would be interested to hear how you migrate a system to different virtualised hardware without disconnecting open sessions. care to elaborate? ... virtualised hardware might be virtual, but its still hardware as far as the guest os is concerned.

Re:Some might argue

[Imagix]

You apparently have not used VMotion on VMware, or Xen has an equivalent technology. Heck, with a sufficiently advanced VMware license, you can have the VM change both hosts, and what disks they are based on without disrupting the guest VM. (I haven't tried the storage VMmotion yet myself). VMware can also hot-add and remove CPU, disk, and memory resources.

Re:Some might argue

[crutchy]

Re: "allows a person or company to "write once, use for years"" ... I can't even run 16 bit programs on Windows 7 (without virtualized "XP mode"). clearly using things for years isn't on microsoft's agenda

Re:Some might argue

[crutchy]

"In 2 to 10 seconds" - i could do the same thing with a toggle switch in a cat5 splitter... 0.05 second downtime! ... i can see where you're coming from though (for vms under the same hypervisor). the problem with your scenario is that you aren't really upgrading anything. you might do it if your guest os has been corrupted or riddles with malware, but otherwise you're still just migrating from one vm to another on the same physical hardware. does it work as well if you migrate to different hardware? ... yes i know google is my friend, but i would rather you back your claim here (if that's ok), and i'm not implying you're wrong; it just seems a bit incredible on face value.

Re:Some might argue

[The Dawn Of Time]

You're pretty clearly not experienced with much if you think virtualization is only for datacenters to squeeze more money out of customers.

Re:Some might argue

[Per Wigren]

Stop hustling him. The topic is how to run several colonels on the same hard wear.

Re:Some might argue

[cloudmaster]

Funny, I prefer my router which runs udev because it's nice to be able to force the order of NICs based on MAC instead of relying upon the order in which they're detected. It's made my iptables scripts a lot simpler.

It's only a pain when you forget it's doing that and spend hours trying to figure out WTF is going on. Not that this has ever happened... ;)

Re:Some might argue

[cloudmaster]

No, I meant initrds which already have all of the modules included so your kernel's not too large for the bootloader to handle. :) But then, I don't know if that's a real problem any more, since I've not built a fully monolithic kernel in probably over a decade.

Re:Some might argue

[cloudmaster]

Right though you may be, it kinda is his personal playtoy. Hence the name. :)

Re:Some might argue

[Kagetsuki]

I've migrated embedded systems and USB installs with dd, but for a real system you should be using something like rsync and partitioning/formatting the target disk in advance. The reason being that with dd it's really a one-shot thing, and dd will copy blank space and random sectors of deleted information etc. so you actually end up wasting time on anything big.

Re:Some might argue

[lennier]

Hi, "hassle" rhymes with "castle"

In what accent? Not mine - here in NZ, "hassle" rhymes with "tassel" and "castle" rhymes with "parcel".

Re:Some might argue

[Anonymous Coward]

Needs to BE fixed,or needs fixing?

Re:Some might argue

[Anonymous Coward]

I say banana, you say bah-nah-nah.

Re:Some might argue

[Thalagyrt]

Uhhhhh?

To put it very clearly: No. They're not flipping from one guest to another. The same guest, moving from one physical piece of hardware to another, without any interruption at all. Why this is mind boggling to you is beyond me - it's fairly simple, actually, once you realize that we have shared storage (y'know, logical volumes you can access from more than one machine - is that equally incredible?) and very fast network links...

Since you're so lazy: http://lmgtfy.com/?q=live+migration

Re:Some might argue

[vux984]

datacentres use virtualization merely to squeeze more servers out of their existing hardware (equals more paying customers) so virtualization for anything else is really just a fad

Lot more uses than that.

unless you use windows in which case the operating system is the single point of failure).

Funny. But bashing windows servers stopped being funny a while ago.

but the simplest approach would be to disconnect the network cable from the old server and plug it into a ready-configured new replacement.

Really? And the 300 database queries and updates that are being executed on the "old server"? They succeed without issue? No database consistency loss either?

The file download that was in progress? Does it continue uninterrupted?

All the logged in users? They're all magically logged in after you plug into the new server? With all the processes and applications they had going in precisely the same state?

Yeah, no... i don't think so either.

i'm aware of tools like heartbeat and pacemaker for automated/remote failover, but i'm not experienced with that.

Or virtualization. :p

Re:Some might argue

[igreaterthanu]

virtualization for anything else is really just a fad, and it has nothing to do with qos or uptime

This is not true.

Suppose you have 8 different tasks (or users on Amazon) that each require the equivalent of one high end machine and you have 8 different high end machines. You could run each of the tasks on one machine and if that machine fails, the task fails. If you instead split each machine into 8 VMs and run one VM per physical machine per task and distribute the load between the machines at runtime; then if a single physical machine fails, the system doesn't go down.

Re:Some might argue

[man_of_mr_e]

I bet if you pulled the hard drive out of a modern version of Linux and hooked it up to a PII, your experience would be highly different.

Re:Some might argue

[crutchy]

i'm not lazy, and you haven't proven anything. i'm fully aware of shared storage; how do you think i could successfully switch a service from an old server to a new one using a cat5 splitter with a toggle switch if the two machines weren't accessing the same data? shared storage has little to do with benefits of virtualisation. my question was how to you change hardware without disconnecting sessions that are connected to the guest os? i'm also aware of fast network reconnections, but if i was in the middle of copying a big file and someone migrated the server through which i was copying from (using shared storage or not), i can't imagine how my copy wouldn't be aborted - it seemed like that's what you were implying you could do with your op. i would be quite happy to be proven wrong. just don't expect me to be fooled by irrespective technobabble. i found http://www.vmware.com/products/vmotion/features.html via your google link and there was nothing that seemed overly impressive and nothing that mentioned maintaining network sessions ("zero downtime to the end-users" doesn't mean it doesn't lose connections - even if its only for an instant).

Re:Some might argue

[crutchy]

you didn't happen to mention any other uses (not very convincing), bashing windows will always be funny (duh!), you not heard of mysql cluster or shared storage, i would be interested to know which virtualisation solution allows migration to different physical hardware without terminating file copies, and all this stuff seems awfully datacenter related... funny that. i'm not an expert on the mechanics of every virtualisation solution out there (my op was an opinion - obviously, but i do use virtualbox), but virtualisation is clearly your only area of expertise.

Re:Some might argue

[crutchy]

lol i didn't actually say that. i said "datacentres use virtualization merely to squeeze more servers out of their existing hardware (equals more paying customers)", which doesn't imply any more money out of each customer.

Re:Some might argue

[crutchy]

... an who would do this other than a datacenter? your example (amazon) wasn't really a good one for proving whatever point you were trying to make, because amazon has huge datacenters.

Re:Some might argue

[hawkinspeter]

If you really want to fork the kernel, then go for it. Hire the appropriate people if you don't have or want the necessary skills. If you haven't got the money for that, make a good business case as to why your ideas are better and get some financial backing.

I'm no expert on ABI, but it looks like there's some projects related to it that are at a higher level than the kernel (e.g. Gnome, KDE etc) which is presumably due to the difficulty of having an ABI on a kernel that supports multiple processor architectures. How would you get the same binary to run on ARM and x64 without drawing a line in the sand and hampering progress?

Re:Some might argue

[Thalagyrt]

In a matter of about 1/10th of a second, you can suspend a VM, migrate its state to another host, resume it, gratuitously arp, and thanks to the fault-tolerant design of TCP, things just continue working. If you knew anything at all about networking you'd understand this.

Instead of being a dumbass and saying "Not possible, no way!" why don't you go install XenServer on two boxes, pool them, and see it for yourself? Or, how about simply not opening your mouth on a subject you yourself admit you know nothing about?

Re:Some might argue

[turbidostato]

" i could successfully switch a service from an old server to a new one using a cat5 splitter"

Can you do it from 1000 miles away?

"my question was how to you change hardware without disconnecting sessions that are connected to the guest os?"

And the answer, again, is "live migration". And if it's needed to be more precise, like this (Xen example):
`xm migrate --live virtualmachine destination-node`

"if i was in the middle of copying a big file and someone migrated the server through which i was copying from (using shared storage or not), i can't imagine how my copy wouldn't be aborted"

Take my word for it: it is not aborted.

Regarding your imagination, well, that's the point: you can't imagine how, others did and implemented it. Certainly you seem to be lazy enough not to look for how this technology works (Google is your friend, there's aplenty documentation about how live migration works with regards of in-memory data) but still rude enough to discredit what a bazillion systems administrators are doing and have been doing for years.

""zero downtime to the end-users" doesn't mean it doesn't lose connections - even if its only for an instant"

I think you should take a bit more time understanding how a network connection really works. I.e.: learn that from its very begining TCP/IP was conceived to work through unreliable lines so surviving a dropped packet here and there is already taken into account in the very protocol.

Re:Some might argue

[crutchy]

what i find amazing is that you can be so offended why what he says. what did he personally do wrong to you that he didn't have every right to do? what way did he mistreat you that was different to anyone else? what do you think he owes you? just because people listen to him doesn't mean he has no right to voice an opinion. i don't think anyone could doubt that he's done a lot for linux, FOSS and the IT industry in genral, and because of that people listen to him, whether you like it or not... also, a lot of software used by companies is still 16 bit. just because microsoft (or you) say its old doesn't mean it should be forgotten. re: "you have ANY leg to stand on when it comes to BC" and the rest - i don't really give a crap. if you hate linux and love windows then good for you. i don't use linux because i hate windows; i use it because its useful for me and because its free. i reckon linus seems like a decent enough bloke who is pretty smart and a lot of people are jealous because you can't buy his credibility.

Re:Some might argue

[crutchy]

you're most convincing statement in your reply was "Take my word for it: it is not aborted.", and only because it would seem from it that you have tried it yourself. now take my word for it: all the rest was bullshit. copying a file has as much to do with tcp as saving a document has to do with FAT32.

Re:Some might argue

[crutchy]

being able to migrate a guest os to different hardware without interrupting a file copy has nothing to do with tcp and everything to do with maintaining the instance of the program doing the copying. perhaps you are more ignorant than i gave you credit for.

Re:Some might argue

[Thalagyrt]

Ok. Let me lay this out for you, since you're apparently a dunce who can't use google.

How live migration works:
1. Snapshot system memory, live, while the system is running.
2. Transmit snapshot to shared storage.
3. After transmit, load snapshot into RAM on second host.
4. Suspend VM.
5. Send deltas of RAM, to bring second host up to date.
6. Resume VM.
7. Gratuitously ARP out so the switches know where the machine is.

The guest is never shut down. It is never aware it has even been suspended. As far as it, and clients are aware, it just experienced additional network latency for about 1/10th of a second between 4 and 6. During that time, its entire state was transferred to different physical hardware. Now, in many cases, this requires identical, or similar CPUs, but that's trivial.

It's not that complex, and if you were willing to do a bit of research, you'd avoid making yourself look like a total dunce.

Re:Some might argue

[Anonymous Coward]

...mysql cluster...

Oh, the horror!

I refer you to technologies like Platespin Protect to further your education.

Re:Some might argue

[crutchy]

i didn't say it was impossible. i even said "i would be quite happy to be proven wrong". the problem with your latest attempt at bewildering me with bullshit is that it would seem possible to migrate a suspended guest os to identical/similar hardware. you would be taking a pretty big risk trying to resume a host on a different machine without any problems, and i'm guessing you would need a volume license for a windows guest for activation. no i haven't googled; its more fun winding you up :)

Re:Some might argue

[crutchy]

if it works :)

Re:Some might argue

[turbidostato]

"all the rest was bullshit. copying a file has as much to do with tcp as saving a document has to do with FAT32."

Of course I made the suposition that you meant copying a file *over the network* (say, scp ./this/file user@remote.host:that/place) because suggesting that there would be problems when copying a file *locally* (i.e. cp /this/file /that/place) is so utterly uninformed as to immediately discard such interpretation.

Obviosly I underestimated your ignorance on the basis that usually a sane person doesn't try to challenge common knowledge on a professional area when not having the slightest idea about it.

Re:Some might argue

[crutchy]

this would of course only be a problem if you *wanted* different guest hardware. one advantage of a virtual machine is that you can have all the same virtual hardware on different physical hardware. i found out a little about live migration (did i hear you say "omg!") and it does seem impressive - failover with incremental memory copying. definitely datacenter stuff. if you design your network right a regular failover cluster can be just about as good though.

Re:Some might argue

[cheekyboy]

try migrating live 5000 VMs from 1 cluster center to another.

we arent talking about little home blogs here but real setups that have 10s of millions of customers accessing the services.

Re:Some might argue

[Guido von Guido]

It's not just fairly simple--it's mainstream. VMware does this, I think Xen does, IBM does, and even Microsoft does.

Re:Some might argue

[crutchy]

Re: "I think you should take a bit more time understanding how a network connection really works. I.e.: learn that from its very begining TCP/IP was conceived to work through unreliable lines so surviving a dropped packet here and there is already taken into account in the very protocol."... what does this have to do with live migration again? "copying a file has as much to do with tcp as saving a document has to do with FAT32" was merely to highlight that while tcp is in there, live migration requires maintaining the copying process (higher level). i don't know where you were going with the networking thing or why you thought i needed to brush up on it. also, would it comfort you if i said i was insane? i like to think a bit more critically rather than blindly accepting whatever someone tells me (including google/wikipedia/microsoft/slashdot) so challenging common knowledge isn't beyond me. i would be happy to be lumped into the same class as copernicus, but of course the sun revolves around the earth!

Re:Some might argue

[turbidostato]

"TCP/IP was conceived to work through unreliable lines [...] what does this have to do with live migration again?"

Quite to the point that of "again", because it has been told to you already (see #37151918). Hint: MAC to IP relationship.

"i don't know where you were going with the networking thing"

Of course you don't.

"or why you thought i needed to brush up on it"

Because you don't grasp what's its relationship with the live migration problem as you don't grasp that copying a VM memory status to a different machine is, logically-wise, the "easy" part (but devil is on details).

"i like to think a bit more critically rather than blindly accepting whatever someone tells me"

That's not critical thinking, that's merely being a smartass.

"challenging common knowledge isn't beyond me"

Here I'l throw you an idea: challenging common knowledge *on a professional area* while at the same time lacking obvious knowledge in such area tends to be quite stupid.

"i would be happy to be lumped into the same class as copernicus"

I bet you never read "De revolutionibus orbium coelestium", I bet you don't know what his real thesis was and why and I bet that your idea about Copernicus comes from his "common knowlege" image. Quite something for someone that says about himself "challenging common knowledge isn't beyond me".

Re:Some might argue

[visualight]

Wow. I would love to discuss this with you at a coffee shop or something. I mean holy shit you're insane (and really really ignorant).

Are you on youtube?

Re:Some might argue

[vux984]

http://blog.inetu.net/2009/03/7-great-uses-for-virtualization/

First link when you search for "uses for virtualization".

Additionally not on that list there is dynamic load balancing between different services.

Lets say you have 11 different low load servers (some database servers, some web servers, your voip mailbox system, requiring different operating systems, different users, different support contracts... etc... eg. the telephony contractor has admin remote access to the voip system). Its all virtualized onto 3 physical systems.

It would be a waste of space and money to have 11 physical servers for them. Most of them idle at 6%-10% of CPU most of the time.

Then one of them gets a load spike... you can live migrate it so that you've got the 6 low load servers on two servers, and the busy one on the other hardware with no downtime.

Three days later, one of the other servers gets a load spike... shuffle them around.

Then hardware fails, so you migrate all 11 onto 2 servers while you get it fixed. Thanks to shared storage, the servers running on the failed unit, are rebooted onto the surving ones with minimal downtime. (See, shared storage is particularly useful in tandem with virtualation... but it solves separate problem.)

you not heard of mysql cluster

What would that do for me here? My mysql database load is minimal... I don't even need one servers resources allocated to it, nevermind a cluster of them. I do still want to reduce downtime. Virtualization solves that neatly. Clustering gets me that as well, but in a tremendously inefficient way considering the load I have.

or shared storage

See above.

i would be interested to know which virtualisation solution allows migration to different physical hardware without terminating file copies

Vmware
Microsoft Hyper-V
Xen

"Administrators can "live migrate" virtual machines between physical hosts across a LAN without loss of availability.
During this procedure, the LAN iteratively copies the memory of the virtual machine to the destination without stopping its execution. The process requires a stoppage of around 60â"300 ms to perform final synchronization before the virtual machine begins executing at its final destination, providing an illusion of seamless migration."

Re:Some might argue

[hairyfeet]

Riiight. Your response is an insult because you have NO ANSWER to all the links I posted showing retailers RUNNING AWAY from linux as fast as they can. Did you read the post I responded to? Crutchy was going /pulls sheet over head/ "Why are you picking on poor wittle linus? what has he done to you? linux is teh leet! It gives you gonad powerz!"

You and the precious community can have little circle jerks and tell each other it is a secret conspiracy by gates and the illuminati, but that won't make you ANY less full of shit. here is some facts for you, like it or lump it...

FACT it has been TWENTY YEARS and Linux has gained NO SHARE on the desktop. it was at 1% nearly a decade ago, guess what it is now? FACT As iOS and Android and Win 7 have proven beyond a shadow of a doubt THE TERMINAL IS DEAD as far as consumers are concerned. What does Linux do? 'they don't know whats leet! We'll put the term on teh desktop and make everyone use it! that'll give teh gonad powerz!"

You can sum up EVERY Linux post by simply going to TM Repo and picking out what standard bullshit excuses get trotted out. oh its the OEMs, oh it retail, oh its everybody's fault but ours! Bullshit, complete and utter bullshit. Waste you mod points all you want but it won't turn your complete and utter bullshit into anything resembling reality, and that's what hurts don't it? that even FOR FREE nobody wants what you're shoveling? When MSFT puts out a turd and your FREE OS gains NO SHARE while an OS with a $1000 barrier to entry stomps your ass? wake up and smell the shit buddy, wake up and smell the shit.

Re:Some might argue

[alevchuk]

Migration is a fundamental feature of OS-level virtualization. Checkout OpenVZ live migration.

Re:Some might argue

[Anonymous Coward]

I can only agree. I have some virtual machines created back in 2000. They run happily on newer hardware - faster. Easy to move, backup and test for upgrade, etc. I have in average 200 virtual machines that I use.

Re:Some might argue

[crutchy]

lol you stupid wanker

Re:Some might argue

[crutchy]

haha i call shill (actually your already a known shill). linux was never designed for "the desktop" or for making money. that it has made any headway at all is merely a side effect. you go have your wingasm but i won't be joining you because your jealousy of linus really stinks.

Re:Some might argue

[crutchy]

re: "first link..." - why would i search when your content to do it for me. so predictable :) tahnkyou for the info and this discussion as a whole has been enlightening. i didn't know about live migration before (i think the latest i heard of is now called "quick migration"). its been fun teasing out the info.

Re:Some might argue

[crutchy]

its funny that tcp didn't feature anywhere in your how it works summary, and it took all of what you said to basically repeat what i had already said in one sentence "everything to do with maintaining the instance of the program doing the copying". thanks for the info though.

Re:Some might argue

[crutchy]

This: "I'm your worst fucking nightmare...a retailer that still believes." sums up all your rage against Linux... its a totally awesome operating system... THAT YOU CAN'T SELL FOR A PROFIT!!! hahaha. poor thing. Re: "WTF does what he did in the past matter now"... i guess the people who invented the transistor don't matter to you either. maybe you should join the hippies and abondon technology altogether you ungrateful worm. Re: "What corp in their right mind would be running 16 bit apps nearly 15 YEARS after support ended? Security hole much?" whoever relies on applications (or the thunking to run them) for security is a moron, and its not a case of corps being in their right mind; they may have no choice if there's no alternative. not every company uses photoshop or autocad that keep forcing you to upgrade; i'm talking about CAM, datalogging, POS, etc. some programs are also good as they are. just because operating systems don't want to support them anymore doesn't mean there is fault with the 16 bit app. if it aint broke don't fix it, and some 16 bit apps work very well for what they do. i'm not a linus kiss arse as much as you clearly despise him. i merely appreciate that he paved the way for the OS that i now freely use. why do you insist that linux must compete with windows? linux isn't about profit or market share or usage. are you too thick to realise. nobody except microsoft shills really give a shit, but its still fun to get you all wound up :) the rest of your post was pretty funny stuff... "cower at the truth". you're a nutcase dude. insults won't make me cower, and i'm not going to treat anything you say as truth because you're a known bullshit artist. not that i mind; keep going because its quite entertaining.

Re:Some might argue

[monkyyy]

thats because in mirco$ofts world they dont release the source

Re:Some might argue

[Anonymous Coward]

Linux isn't an operating system.

Re:Some might argue

[Thalagyrt]

It's amazing how content you are with ignorance. I'll just go ahead and add you to my "Never met a smart Australian in my life" list and leave it at that.

Re:Some might argue

[crutchy]

since you think you're smarter than everyone else, i really have no choice but to be content with your ignorance :)

Re:Some might argue

[diegocg]

ALL of the major OSes, MSFT Windows, Apple OSX, BSD all flavors, hell even OS/2 has an ABI.

Linux does have an ABI, it's called "syscalls". Drivers are a completely different beast. You are mixing both.

overhead for an ABI would be trivia

Here is where you show that you don't know what you are talking about. Why would an ABI have any overhead at all?

the fucking driver mess

Despite all the people who claimed, for decades, that making hard to develop binary drivers would be bad for Linux, it turns out that the Linux driver support is quite good. These days I don't even need to care about the hardware specs of a computer when I put Linux on it. And, unlike it usually happens with binary drivers made by hardware companies, Linux still supports old hardware. Try using Windows 7 64 bits in a old computer with an old graphics card and an old scanner.

Linux will stay stuck at 1%

Linux owns the smartphone and server market.

Re:Some might argue

[Thing 1]

Way to encourage helping.

Re:Some might argue

[Ben Hutchings]

Try migrating a running system with dd. Many VM systems can do live migration, with a fraction of a second of down-time.

Re:Some might argue

[Anonymous Coward]

The lack of a stable driver ABI or API protects linux from buggy closed-source drivers which no one can fix. If you can't write an open-source driver good enough to be accepted upstream, then no one can even try your driver with the latest kernel release. Well NVIDIA manages I suppose.

I don't always agree with linus, but frankly he doesn't do that much coding anymore, he just to has push back against crap code, and he does a great job, and the quality of linux reflects that. You even agree that it's a good OS, except for your pet feature - and everyone has a pet feature, and that can become a huge problem if there isn't someone like Linus in control.

Re:Some might argue

[hairyfeet]

Cowards, that is ALL you and your community really are. You hide behind your Tux blankies and if anyone dares to say anything other than the circle jerk "Gee isn't Linux swell? It sure is Biff, and RMS' cock tastes like candy!" you go "Oh Noes, it is not teh true! our precious is leet! it gvives teh gonad powerz! you are teh secret M$ Ninja!" which hey guess what? Just like all the other horseshit you sling, has a trademark because you don't even have any originality to your shit. What a fucking surprise from the OS that rips off Windows and OSX, you know,the two OSes that are actually functional?

Hiode behind your blankie coward. i provided over ONE DOZEN LINKS showing fortune 500 retailers RUNNING AWAY from your POS as fast as they can. Why? Because YOU DO NOT LISTEN TO USERS and instead hang onto your "CLI haz gonad powerz! it makes us teh hackrz!" horseshit. Meanwhile your little rant is such a fucking joke they already have a trademark for it

Re:Some might argue

[grimharvest]

So you just flat out don't like Linux, but you decide to read a Linux thread anyway, don't agree with Linus on something, then go into a rant? Why are you even here?

Re:Some might argue

[Anonymous Coward]

" i didn't know about live migration before"

Do you not do any research of any kind? "Live Migration" has been touted as been one of the largest features for virtualization for a while now. Every marketing team, review site, and feature list has it front-and-center.

Being in the tech industry and never hearing of "Live Migration", is like being an average consumer and never hearing of a car. What, they don't use horses any more?!

Re:Some might argue

[Bengie]

An OS live migration does a live copy of memory and CPU state. From the view point of the OS, nothing has changed. Nothing get's stopped or interrupted. The OS never sees anything happened, the client never sees anything happened. It's 100% transparent to all users in the system except the host OS. A small burst of packetloss during the hand-off window may happen, but packetloss is to be expect at some point in any networking system.

To be fair, Thalagyrt has been more correct than you, and quite patient with your lack of common knowledge. You pick one his one stumbling of words, which were caused by your inept knowledge on the subject coupled by your refusal to do basic research, then attempt to tear him down.

What Thalagyrt meant, is the only thing the end user may see is possibly minor packet loss during the final switch over. There is a very very small window of no response during the hand off. "fault-tolerant design of TCP" just meant the TCP stack would handle re-sending any lost data. Yes, the point is moot because every program must assume packet loss, but I knew were he was going with it and I've made the same mistake only 101 times myself. Don't focus on the fly in his distilled water, when you're drinking chemical sludge.

Re:Some might argue

[Bengie]

obvious troll is obvious

Re:Some might argue

[Bengie]

It would be, but application are meant to run within an OS, not across several. Also, as a system admin, you don't manage applications, you manage systems.

By server, I don't mean the physical hardware, unless stated, I mean the actual OS instance. The hosted admin manages the web servers, the db admin manages the db servers, the storage admin manages the SAN, and the systems admin manages the physical server. There is very little overlap.

The hosted admin doesn't need to know which system is web server is running on. The storage admin doesn't need to worry about anything except making sure the SAN isn't getting overloaded. The Systems admin doesn't need to know what runs on the server or where the data resides, he just needs to know how loaded the server is and shuffle around the load to other servers if need be.

The logical hierarchy just makes OS virtualization a better choice.

Linus is right

[mfh]

The shift towards virtualization represents a further shift in control away from each person towards a reliance on the honest of others.

Re:Linus is right

[Anonymous Coward]

If the layer below is as transparent as the higher layers, how is that the case? Virtualization is just a layer of abstraction.

Re:Linus is right

[Luyseyal]

Suuuuure, you have 2 CPUs, Mr. Guest OS. *wink*wink*

-l

Re:Linus is right

[0racle]

If I run Linux as a host and FreeBSD and Windows in kvm or VirtualBox, to whom have I given up my control too?

Re:Linus is right

[MightyMartian]

WTF? I've built three production Linux KVM servers now. Other than relying on the KVM team (backed mainly by Redhat), I'm not relying on anybody else. And if Redhat is a problem for you, then you've got bigger issues than the KVM virtualization modules in the kernel.

Re:Linus is right

[Anrego]

If you are talking virtualization in it's relationship to cloud computing, then I agree.

Virtualization on your own hardware though... not much difference from the current state. It's just another (sometimes open source) piece of software we learn to trust (along with all the other software we use).

Re:Linus is right

[Anrego]

* its

Please forgive, it's Friday :(

Re:Linus is right

[arth1]

In the case of VirtualBox, that would be to Oracle.

Can you say with certainty that they will support Windows 9 guests or Linux 3.4, or have you given up the control over upgrades to them?

Re:Linus is right

[Jeremi]

If I run Linux as a host and FreeBSD and Windows in kvm or VirtualBox, to whom have I given up my control too?

The Chinese government, of course.

(ducks)

Re:Linus is right

[Baloroth]

VirtualBox is GPL (extensions aren't, but those aren't needed for core functionality), so really not much control is handed over at all (if Oracle refuses to offer support, the project can be forked.)

Re:Linus is right

[0racle]

Can you say for a certainty that your current hardware will support Windows 9 or Linux 3.4? Certainly now. Are you CERTAIN unreleased future OS's will support your current hardware?

No you aren't, you can't be. Oh they probably will, but then again, VirtualBox and kvm will probably support the next few releases of the major OS's as well. So again, what have I given up, other than 2 keyboards and at least one mouse of course? There are many tools out there to convert virtual disks from one format to another so moving data to another VM package (if you just had to keep it local to that VM for some reason) is easy, and much cheaper than buying adapters for hardware interfaces.

Re:Linus is right

[jellomizer]

I am not following you?

Virtualization is just taking your own Hardware and sharing some resources and partitioning others so you can run Multiple OS's on your hardware...

It is not much more of a jump of putting your faith on the OS who will manage your hardware, and give you an interface level to the hardware.
Virtualization you can think of it as a OS for your OS.

The main value in Virtualization is from an Old computer science process... Keep you hardware running at high utilization doing something productive. As time went on people started to use separate servers running at 5% utilization. Because they needed all their Apps to run separate of other apps. Or groups of the same app running with different configurations. So now we virtualize on the same hardware to say for example to get that CPU 5% utilization to 50% and running 5 VMs sure there is some bloat going and it isn't efficient. But it is more efficient then having 5 servers running at 5% CPU utilization wasting power.

Then you have the more enterprise ready features of a virtualization where say your VM is on a SAN storage then you can swap hardware in real time. Or move from one Server to an other easily and without having to reconfigure your hardware settings for the new PC.

I hope one day Linus or RMS states they like the taste of poo, just to see how many nuts out there will just blindly go along and do it because these guys have an opinion and because they have done or are doing great things they must be always right.

Re:Linus is right

[fnj]

Forgiveness is the fragrance that the violet sheds on the heel that has crushed it.
- Mark Twain

You are forgiven. Now pick me and put me in some water.

Re:Linus is right

[arth1]

Can you say for a certainty that your current hardware will support Windows 9 or Linux 3.4? Certainly now. Are you CERTAIN unreleased future OS's will support your current hardware?

No, but I am certain that I can buy hardware that is compatible.
With a VM, there are no such guarantees, because there's a lack of purveyors of virtual hardware. It's pretty much only the virtual machine vendors, so yeah, you have to trust them to upgrade whenever necessary.

There are many tools out there to convert virtual disks from one format to another so moving data to another VM package (if you just had to keep it local to that VM for some reason) is easy, and much cheaper than buying adapters for hardware interfaces.

Switching to another VM isn't always an option either. So far, none of the VM vendors have provided drivers compatible with Gnome Shell, for example. (Not that gnome shell is essential, but it proves the point by example, I should think.)

Re:Linus is right

[element-o.p.]

How do you figure? Maybe I'm just not the sharpest tool in the shed (nor have I ever claimed to be) but I really don't see much difference between trusting Intel/AMD/Motorola/etc. to be honest in what they put in the chipset and in trusting Citrix (Xen)/VMWare/KVM/whatever in what they put in the software. If anything, I probably have a better chance of detecting goofiness in FOSS software than I have in a chipset (not that I'm likely to detect anything in either one, but at least I *can* look at the source in a FOSS project if I want).

Re:Linus is right

[dissy]

The shift towards virtualization represents a further shift in control away from each person towards a reliance on the honest of others.

And who exactly are you to tell me what I can and can not do with my own hardware?

If I choose to run zero, one, or a hundred OSs on my virtualized hypervisor on the hardware I purchased, what right do you have to say it is a bad thing?

If anything, you dictating that I shouldn't virtualize my own hardware because it is evil, is what would be taking control away from me. Virtualization gives me control. MUCH more control than the hardware itself can.

And what are you going on about relying on the honesty of others?
How does other peoples honesty even come into play when I choose to use my hardware in one way versus another? Other people aren't even involved, let alone their honesty!

Re:Linus is right

[Lunix Nutcase]

(Not that gnome shell is essential, but it proves the point by example, I should think.)

It doesn't.

Re:Linus is right

[ChrisMP1]

I hope one day Linus or RMS states they like the taste of poo, just to see how many nuts out there will just blindly go along and do it because these guys have an opinion and because they have done or are doing great things they must be always right.

It's a good thing you put that at the end, or most people would never read past it.

Re:Linus is right

[next_ghost]

I'm pretty certain that Windows 9 won't support any of my current hardware as 3 of my 4 machines are already too old to run Windows 7 and the remaining one can run it just barely. On the other hand, I'm even more certain that Linux 3.4 will run all of my hardware because my oldest machine was made before Linux 2.2 was released and it's still working fine running Linux 3.0. There's actually much higher chance that all of my hardware will die before Linux 3.4 comes out than that Linux developers will drop driver support for any of it by then.

Re:Linus is right

[crutchy]

i think the op was referring to placing trust in datacenter companies rather than an inhouse it department or local support company, not running virtualised servers on your own hardware - what would be the point of that? do you really think microsoft developed hyperv for you to run it on your own hardware (or office365 for that matter)? its getting more and more cloudy everyday.

Re:Linus is right

[dissy]

do you really think microsoft developed hyperv for you to run it on your own hardware

Actually yes, in fact all of the virtualization companies/groups release their software to run on your own hardware.
(In fact if I ran it on hardware that wasn't mind, I'd likely be brought up on some charges :P )

The main point there is whomever installs virtualization software generally does so on hardware they own or are responsible for.

Most of the servers I maintain at work have virtualization software on them, and run many instances of servers.
I have a single server at home also running virtulization software so I don't need a server farm to play around with other software, while still providing stability on the VMs I do use as servers.

Virtualization is no more than cloud than windows is the only operating system.
The cloud uses virtualization, but so do many many other things. Anyone jumping to assume mentioning virtualization automatically means cloud is really showing they don't understand either technology.

Replace 'hyper-v on your own hardware' with 'Windows on your own hardware'
Just because some online services and websites decides to run windows under it, does not mean one might not want to do the same at home. (Although choosing Windows as an example there was not wise of me)

Re:Linus is right

[PyroMosh]

I agree with you. Calling it "evil" is stupid. Linus is an insufferable prick. Yada yada.

That said, the poster has every right to tell you whatever he or she wants. They're not stopping forcing you to do anything, just getting on a soapbox and spouting off about nonsense.

Keep things in perspective. Linus and those that agree with him are spouting off an opinion, they're not a thugs coming into your home/business to prevent you from using your hardware in an unapproved way.

Re:Linus is right

[Thalagyrt]

Oh hey, look, let me Google this one for you too:

http://lmgtfy.com/?q=Hyper-V

See, this is part of Windows Server, which you install on your own hardware, just like XenServer, or VMware, or well, any other operating system. It's not a hosted platform that you can't control that runs somewhere else.

While I have you here, I really would like to know two things:
1. What the fuck are you smoking?
2. Can I have some? Must be some good shit.

Re:Linus is right

[kiddygrinder]

what a crock of shit, are you saying that buying a virtual machine somehow makes the provider less honest? if your machine is in someone else's datacentre you better fucking trust them, virtualisation has nothing to do with it.

Re:Linus is right

[crutchy]

i didn't imply virtualisation meant cloud, and you took my comment out of context. obviously virtualisation is meant for datacenters and running multiple servers on the same hardware. what the op was getting at is that (non-datacenter service provision) companies who virtualise using datacenter services are putting trust in people who are more remote to them than the traditional in-house IT dept or local service company. virtualisation may not be the same as the cloud, but the two are very closely related; the cloud wouldn't be economically viable without virtualisation. software as a service doesn't depend on virtualisation (just your typical lamp config) but hosting business servers remotely is the domain of datacenters and until virtualisation came about they could only accommodate a fraction of the clients they can now. you may manage multiple virtualised servers in a datacenter environment, but the op was talking about the rest of the world losing control of their data to these datacenter companies (so perhaps in a way he was targeting you).

Re:Linus is right

[crutchy]

unless you are a geek/hobbyist or you are managing a datacenter, what would be the practical use for virtualisation? i'm the former and i use virtualbox to to run windows on one of my linux boxes. i don't smoke and you microsoft shills appear to be the ones with clouded minds (no pun intended). you seem to think just because i understood where the op was coming from that i don't know anything about virtualisation. While I have you here, I would really like to know how it feels to be a moron?

Re:Linus is right

[Thalagyrt]

I like how you assume I'm a Microsoft shill just because I know what Hyper-V is. You see, I do manage datacenters for a living. My virtualization, however, is entirely XenServer, so, hey, again on the assumptions? See, your problem is that you talk authoritatively on subjects you openly admit you know nothing about.

Re:Linus is right

[crutchy]

wow! you think i'm authoritative? lol you poor thing. perhaps you should get out of your datacenters a bit more often (joke). i'm simply having an online discussion, and so far its been quite entertaining (as always on slashdot). apologies if my assumption was off... there are a lot of shills promoting all sorts here (as you're no doubt aware); i figured i probably had at least a 20% chance of getting a bite :) you're right about me not having experience with virtualisation, but it doesn't mean i'm totally ignorant... and if i don't know i can always just wing it. i find its a good way to learn because there's always people here that will jump at a chance to prove me wrong and demonstrate their superior knowledge. google is good, but it doesn't know everything and i still have to wade through commercial shit (even wikipedia's page on live migration is pretty disappointing).

Re:Linus is right

[Anonymous Coward]

"And if Redhat is a problem for you, then you've got bigger issues than the KVM virtualization modules in the kernel."

Namely, Redhat.

Screws are evil

[Anonymous Coward]

Because I'm used to working with a hammer.

Linus is not a god, just a guy, with his own prejudices.

Re:Screws are evil

[Anonymous Coward]

Still, he's right. Probably not the best person so say it, tho.

Why can't I suspend an entire app, TCP stack state, files, and all?
Why can't I, in that state, dump it, then restore it?
Why can't I migrate said dump, including the IP address (that I would have bound to the app instead of the machine), to another server? Then unsuspend it.
Why can't I do that seamlessly?

Virtualization is an ugly band-aid. But it's a band-aid for real shortcomings in current OSes.

Re:Screws are evil

[vux984]

Virtualization is an ugly band-aid. But it's a band-aid for real shortcomings in current OSes.

And solving those shortcomings would lead you to separating the OS into 2 layers akin to a hyper visor and OS... and you'd be right back where you started.

And someone would still need to run a *nix server along side a Windows one, or find they need to run two different distros or two different kernel versions... and still need traditional virtualization to do it.

Re:Screws are evil

[VGPowerlord]

Why can't I suspend an entire app, TCP stack state, files, and all?
Why can't I, in that state, dump it, then restore it?
Why can't I migrate said dump, including the IP address (that I would have bound to the app instead of the machine), to another server? Then unsuspend it.
Why can't I do that seamlessly?

Because networking is bound by factors external to the app, not the least of which that IPs are machine bound, not app bound. Or did you forget that a router somewhere needs to know where that IP is routed so it can route packets to you?

Re:Screws are evil

[Anonymous Coward]

wooosh much?

Virtualization is a *tool* to let you do a couple of things.

1) hardware consolidation for persnickity apps and scaling from low load to high.
2) jack up an install and put it back to 'good' state fairly quickly.

The first one lets people run '5' boxes in 1.
The second lets you fix things quickly without having to worry about that crap uninstall program someone didnt bother to test.

Thats it. Your example is a shortcoming of virtualization implementation...

Re:Screws are evil

[Morth]

That's not different from moving a VM from one host to another, the router needs to change its ARP table in that case as well. There's no reason the IP couldn't be app bound instead of machine bound, if you're willing to assign one IP per app (IPv6 would help). But we're lacking the tools to migrate an IP from one host to another, including the TCP sessions. That's a software problem though, and could be implemented if there's demand for it.

Re:Screws are evil

[timeOday]

Read the article. Linus has accepted both KVM and Xen into the kernel, it talks about why he and some other guru think KVM was managed better and is a better implementation.

Let's not confuse two completely different things: if Richard Stallman said something was "evil," it would mean he was morally opposed to it and wouldn't touch it with a 10 foot pole. But Linus calling hypervisor virtualization "evil" just means he'd rather work on hardware, but hey, you want virtualization, go ahead and take your pick of the ones Linux provides.

Re:Screws are evil

[Anonymous Coward]

VMs - it's just a wee bit more control than key apps in their own chroot but is it really that far off in concept.

Re:Screws are evil

[Bucky24]

Because networking is bound by factors external to the app, not the least of which that IPs are machine bound, not app bound. Or did you forget that a router somewhere needs to know where that IP is routed so it can route packets to you?

It's theoretically possible. Just have multiple IP addresses pointing at one machine, then have some sort of modified network daemon that splits off each IP to an app.

In all honesty it would work just like the current port system works, but using an IP instead of an IP:port. It would be kinda wasteful of IP addresses, but with IPv6 that might not be so much of an issue.

Re:Screws are evil

[simcop2387]

Depending on the virtualization technology involved the router may not need to know. I know with kvm each one has a unique mac address that i've got bridged to the real nic. now it might confuse a switch for a bit when it doesn't know where it is, but so far that hasn't appeared to be much of a problem.

Re:Screws are evil

[Anonymous Coward]

Linus is on the right track. Hypervisors are primarily used to work around limitations of operating systems, such as poor isolation, poor security, poor privilege separation, poor storage management, lack of multiple administrative domains, lack of server migration, poor support for legacy software, inability to run Windows executables, etc.. If the OS can overcome those limitations - or if the limitations become irrelevant - then a hypervisor is no longer necessary.

Re:Screws are evil

[The Dawn Of Time]

And let us not forget that the word "evil" actually has no applicability to this situation in any way, shape, or form, but nerds love watering that word down to make their opinions seem more important.

Re:Screws are evil

[Anonymous Coward]

Because I'm used to working with a hammer.

Linus is not a god, just a guy, with his own prejudices.

Linux may not be a god, but he has provided humanity more in his lifetime than has anyone else. His work has touched more people, more often, and has been extended to more generations than nearly all of the history of humanity can account for in others.

The Linus invention [Linux] was a world wide effort; and it has already enabled, as much as did the discoveries of the technologies that enabled water to drink, air to breath, fruit, roots and vegetables to grow together with construction documentation [source code] and without charge; so that individuals, anywhere located, could use it, if only the would take the trouble.

What Linux did was bring the technology needed to participate in our "knowledge storage and information gathering" society to everyday people and mankind in general. That result has challenged nations, governments and injustice around the world. It has brought to light many wrongs which need fixing. The cloud ideas seem destined to corrupt those benefits and, little by little, to disconnect the masses from the technology knowledge that independent progress depends on!
The triad of human rights found in the Declaration of independence suggests removal from ones own control of one's own destiny could once again "allow to hide in secret and mask in user fees and insider's only parlance" the activities of "the greedy and the corrupt ors" and make more difficult discovery necessary for independent people to compete against the established.

Virtualization is expensive in compute ressources

[drolli]

but its cheap in human resources since it is the ultimate reuse of code.

Good for beginners

[Anonymous Coward]

Virtualization is good for new junior programmers learning how to program firmware, sinceeany low level calls can not really destroy the real hardware, since protection can bee built right in.

It's a crutch, but since we have a generation of programmers who can't do "the hard stuff" becuase "java does it for them", its certaintly good to have around.

Re:Good for beginners

[cowboy76Spain]

Sorry sir, but you are in the wrong thread.

Fortunately, nothing of value has been lost.

Re:Good for beginners

[loufoque]

That's not virtualization, that's emulation.

"Hardware virtualization"?

[smoothnorman]

Isn't Hardware _realization_? and/or if the hardware is virtualized then isn't it done with software? not "real hardware"? ...ok, i admit it, i'm lost. someone smarter than me, -you there-, some examples please. (which need not necessarily involve automobiles)

Re:"Hardware virtualization"?

[Coren22]

You go out and buy a Dell blade cluster, it contains 16 identical blades with 2 sockets of 6 cores each and 24 GB of RAM. You hook this blade cluster to an Equalogics array for storage. Install ESX on 15 servers, vcenter on 1, now you can install whatever servers you need, and they are entirely fault tolerant.

Oops, there is smoke coming from one of the ESX systems, and it seems to be unresponsive; vCenter detects the failure and moves the virtual machines without downtime. I don't know what crack Linus is smoking, but Linux cannot do this with very many services.

Re:"Hardware virtualization"?

[johanatan]

The guest OS sees a virtual representation of the hardware which is essentially written as an application that runs inside the host OS and consumes the actual hardware resources (as any application would) on behalf of the guest. See: http://en.wikipedia.org/wiki/Virtualization#Hardware

Re:"Hardware virtualization"?

[Anonymous Coward]

What happens if the vcenter system goes down?

Re:"Hardware virtualization"?

[ArsonSmith]

an entire server for vcenter? Eat your own dog food!!

Re:"Hardware virtualization"?

[eharvill]

Everything retains its current state. You can also manage VMs by directly logging into each host. There is plenty of time to rebuild vCenter if you need. vCenter is a little more critical in View and Lab Manager environments though. But there is also a product called vCenter Heartbeat to handle HA for a vCenter instance.

Re:"Hardware virtualization"?

[Anonymous Coward]

12 cores is a lot of cores to have for only 24GB of RAM. Our standard virtualisation platform is Cisco blades with 12 cores and 374GB of RAM. And vCenter can't move guests without downtime if the host is already unresponsive - vMotion will only work with cooperation between the host you are moving from and the one you are moving too. VMware HA will restart guests on a different host, but there will be downtime.

Re:"Hardware virtualization"?

[rubycodez]

if your machine/software corrupts the shared storage without vmware knowing about it, oops time to go back to a snapshot, hope its not corrupt too. Shared nothing service clustering with proper transaction based replication is far more robust, trivial for the major business applications like databases, email, middleware, file sharing, authentication. I've worked with vmware for a decade, it's wonderous stuff but its not fullproof and not always the best solution. It is mostly used to make up for MS-Window's lack of being a real OS.

Re:"Hardware virtualization"?

[swb]

My experience has been that it's a little easier to run vCenter outside of the cluster it manages. It works to virtualize it, but there are times where it is a nuisance.

What's great if you can get away with it is to run vCenter on a standalone esxi box. You get the independence from the cluster and you can run a backup dc in a separate vm, so even if you lose the whole cluster you at least have something you can work with to get it up.

Re:"Hardware virtualization"?

[Anonymous Coward]

But whoa, what if it could? What would it take to make that work? Quick, someone code this. I need my minecraft server to stay up no matter what!

Re:"Hardware virtualization"?

[smoothnorman]

So far you have got the closest to make some sense of this for me (and, mind-ya, i blames myself) but "hardware which is essentially written"? software is 'written'. So (trying to interpret this) you're saying that in Hardware virtualization the virtual hardware is actually software, yes? So how may we distinguish between virtual hardware and virtual anything else if it remains a software construct? It seems that *all* virtualization is is just a fib to an operating system that here is something that appears to be a familiar service and is actually an emulation of that service. Note that it's ...probably... the terminology that i'm having difficulty with; not what may be done with some dell cluster hardware. thankee.

Re:"Hardware virtualization"?

[Bucky24]

Likely there's a secondary and even a tertiary server for the control daemon itself, and likely multiple mirrored clusters as well.

Re:"Hardware virtualization"?

[johanatan]

Sorry, bad wording on my part. How about this rephrasing: The guest OS sees a virtual representation of the hardware that is provided by an application [i.e., software] running inside ...

So, yea. It is a software representation of hardware.

Re:"Hardware virtualization"?

[Anonymous Coward]

Think LPAR's (Hitatchi Data Systems does this for x86 although its really more of a mainframe thing).

Re:"Hardware virtualization"?

[ArsonSmith]

I run two sites with clusters in both sites, then use vCenter heartbeat to keep backup vCenters for each site.

Re:"Hardware virtualization"?

[Lennie]

I assume ESX saves the state of the VM on the shared storage ?

The problem I see with that is that if the system crashes, you don't have the state at the time of the crash. At best you have the state that's very recent but is prior to the crash.

So you are creating a time warp for your application. I would be very careful with that.

Re:"Hardware virtualization"?

[Coren22]

Yes. We have another cluster where they tried to make it a VM, the power went out and when everything came back, it was actually quite difficult to bring everything back. Having a seperate vCenter server prevents that issue, and it has enough horsepower to run any other stuff that we want to throw on it.

hypervisors are a necessary evil

[arkhan_jg]

VMware makes a hetrogenous environment far, far easier to deal with - we have some 70 odd servers running on 5 physical servers. It makes it much easier to single-task a given server/VM, spread the load without having to invest far more in server hardware, and allows having backup/redundant servers to allow for patching/upgrades of servers with much more minimal effort.

While in-OS virtualization is great if you only require a single OS to do everything; but if you have hetrogenous servers to handle different tasks for different clients - i.e. AD/exchange for the windows users, linux for the webservers and network infrastructure etc, then hypervisors are frankly essential for sysadmin these days.

Re:hypervisors are a necessary evil

[h4rr4r]

Not at all. Put all the linux boxes in openvz or something like it and windows servers can go into their hypervisor until MS smartens up.

Then you just need the linux boxes all on one set of hardware and windows on another. Vmware has other attractions, this is not really one of them.

Re:hypervisors are a necessary evil

[BagOBones]

Accept now you need a Linux Vitalization admin and a Windows Virtualization admin.. You have just doubled the number of "platform specific" gotchas you need to learn, plus unlike having a single VMWare cluster you now have two clusters, probably increasing your hardware cost per vm by needing both platforms to maintain a proper save resource overhead to handle failures.

Re:hypervisors are a necessary evil

[h4rr4r]

Good point.

Any linux admin can probably be taught the "Monkey see, Monkey click" windows stuff though.

Re:hypervisors are a necessary evil

[Anonymous Coward]

Try working in a large company these days.

Every god damned thing has to be a VM, even IO intensive applications such as network discovery tools that ping/ssh/wmi/etc 80,000 servers nonstop. Tell me how good that runs on VMs..... The company I work for is just retarded about this. VMs fail and trip up in all sorts of odd ways. We can't actually snapshot or use any VM features, it's purely for *their* use even though I'm a lead developer and understand the system better than the low-on-the-totem-pole admins.

Please stop giving large companies tools like this. They are far too incompetent to handle them properly. I've manually staged the same identical installation by hand over 40 times when I could have done one and made an image. Their policy is asinine..... Oh and I make six figures as a specialist who now wastes 70% of my time doing basic repeatable tasks on their VMs that I could automate in one day. Oh and the first digit of my salary is not a 1 either. Talk about wasteful.......

Re:hypervisors are a necessary evil

[MightyMartian]

And can be hired to replace the Windows admin who gets palpitations at the site of a CLI.

Re:hypervisors are a necessary evil

[Anonymous Coward]

Which is why my workload is always double that of someone who does only Linux or only Windows. I can work with both. I get paid approximately the same amount of money with more work and more responsibility.

Re:hypervisors are a necessary evil

[Terrasque]

Any linux admin can probably be taught the "Monkey see, Monkey click" windows stuff though.

I take it you've never actually administered a windows system. It's more complex than that. Much more complex. In fact it's mindwarpingly complex, and for all the wrong reasons.

Things are done in not only inconsistent ways, but in many cases downright insane ways. And you almost always have to work through the "friendly" GUI, that have more of a goal to look shiny, rather than being helpful. It's like putting together a tiny, mechanical watch while wearing large, fluffy mittens.

The only way to operate such a contraption without losing all hope and sanity is to treat it like a black box with magic rituals attached. If you try to find the logic in it you'll find yourself wearing your pants on your head.

Once you've seen the light, it's hard to go back to stumbling around in the darkness. And if you dare light it up, you'll see the horrors that lurk there.... And despair.

Re:hypervisors are a necessary evil

[tnk1]

As a Linux admin, I mostly agree with the sentiment, but just because Windows is a GUI doesn't mean it's easy. The amusing thing about Windows is that they have managed to complicate the hell out of server software with all the damn GUI control panels and wizards. Yes, they're pretty, but that doesn't mean you know what everything does.

Re:hypervisors are a necessary evil

[eharvill]

Implying that Linux admins are superior to Windows admins? I've seen more than my share of Linux "admins" that rely on X-windows as well. A good sysadmin is a good sysadmin, regardless of platform.

Re:hypervisors are a necessary evil

[Anonymous Coward]

HetEEEEEErogenous, got it?

Re:hypervisors are a necessary evil

[Locutus]

but I remember a day when the corporate OS was capable of running finance, engineering, sales, and shipping all on the same box and with up times you could easily live with. Then along came Windows and it was like freak'n Tribbles because the OS failed so much they where putting one service on one box/OS. I've already seen virtualization getting praised at Windows shops because of how much hardware it can save. A side benefit is that now they can throw in a LAMP stack without asking and save some bucks because budgets are tight.

LoB

Re:hypervisors are a necessary evil

[Anonymous Coward]

If you knew half as much, and were one fourth as competent, as you claim; you would automate it anyway.

Also, the amount of ones salary is rarely a appropriate indicator of ones ability, and more a indicator of the stupidity of the organization.

Re:hypervisors are a necessary evil

[unencode200x]

Can you give some examples? I find I can do anything needed to manage hundreds of Windows boxes through a CLI (using utilities or powershell). Have you heard of Windows Server 2008?

You're right, though. *PROPERLY* managing a Windows network requires a lot of skill, learning, and dedication, just like anything else.

Re:hypervisors are a necessary evil

[Terrasque]

2008 is a bit newer than what I'm used to. 2k3 is the one I've had to deal with (and is still active at work.. Together when a MSSQL 2000 server... *sigh*). And that is a real PITA to work with.

Re:hypervisors are a necessary evil

[unencode200x]

Agreed on 2003. We still have a few of those around and try to squash them as we go. We actually have one MSSQL 2000 box that is doing some crazy heavy lifting given it's age. Luckily the project to update it was approved and we're working on doing that. It's actually been relatively easy to get it up to MSSQL 2008 R2.

I agree

[MpVpRb]

It's hard enough to get stuff to run reliably when you are dealing with real hardware.

Adding another layer just increases the number of dark corners where bugs can hide.

Re:I agree

[Anonymous Coward]

It's hard enough to get stuff to run reliably when you are dealing with real hardware.

Adding another layer just increases the number of dark corners where bugs can hide.

Conversely, a virtualization environment presents a somewhat "neutral" hardware profile to installed OSs. This makes it useful for installing legacy software on new hardware.

Re:I agree

[0123456]

Conversely, a virtualization environment presents a somewhat "neutral" hardware profile to installed OSs. This makes it useful for installing legacy software on new hardware.

And adds a new load of bugs in the process.

When I was writing PC emulators years ago there were a lot of obscure bugs in the emulated applications when the fake hardware we gave it didn't quite work the same way the real hardware did and there was no way to emulate it precisely.

For example, suppose you have Linux running with an ext4 filesystem that's emulated by a disk file on a real Linux system using an ext4 filesystem on RAID. Do filesystem barriers work?

Re:I agree

[MightyMartian]

That's not been my experience. In many respects virtualization simplifies things. Unless you're dealing with paravritualization, generally you're dealing with a much more "dumbed-down" environment; some limited number of emulated NICs, video cards, mass storage controllers, bus controllers and the like, and these are almost inevitably taken from widely-understood hardware. Yes, there could be bugs lying in wait within the virtualized drivers, but then again, it's not like real hardware has any lack of onboard controller bugs and driver bugs.

To put this another way. I've been running the same Server 2003 and Linux servers under the same KVM host server for two years now, and I have had no more downtime then when I had them on real hardware.

Re:I agree

[BagOBones]

Hardware failures in our VMWare environment tend to go almost completely um-noticed by server admins and en users... We have had network cards fail, Host power loss, and ram failures... The VMWare brings the VMs back on line so fast that there is rarely an alarm on the service, instead just something we need to fix on the back-end.

This goes extra for storage where we regularly shift VMs between SANs by different manufacturers using storage VMotion, LIVE.

Re:I agree

[Anonymous Coward]

A worse example for me is a number of open source operating systems being developed on qemu before ever touching bare hardware/real bios. These systems are listed as supporting anything back to i586/i686 processors, only due to never having run on 'real' hardware that old they have no workarounds for the gotchas of real errata/bios from that era, and thus will often hang/crash on real hardware from that era. Result of this is that I am unable to use other open source operating systems besides *BSD/linux because the software doesn't support the hardware it's claimed to (Even if it's the same chipset as qemu they're running against, it may not 'in reality' have been set up the same way.)

Re:I agree

[sarhjinian]

Yes, it works. Never mind the filesystem, you'd use a real hypervisor on real hardware with real operating systems designed for real virtualization, with real drivers that all work really, really well.

There's a big, big difference between, say, Xen, ESX or Hyper-V running your enterprise apps, and, say, getting Leisure Suit Larry running on Bochs. It's really, really nice to be able to move a running Linux guest with a few hundred users over to another server without a hiccup.

Heck, even at the workstation level it all works very well. I'm sure that crazy old DOS stuff that bangs quirks of a vintage AdLib won't work, but modern operating systems handle virtualization well. Heck, they're designed for it nowadays.

Yes, you can do this sort of thing on a "proper" OS and with applications that can be safely jailed and that support clustering, but sometimes, in the real world where apps and operating systems and people suck, it's easier, quicker and cheaper to use VMware or whatever.

Re:I agree

[0123456]

Yes, it works. Never mind the filesystem, you'd use a real hypervisor on real hardware with real operating systems designed for real virtualization, with real drivers that all work really, really well.

In other words, if you don't go to a great deal of trouble to ensure that you're running the right software on the right hardware, you risk corrupting your virtualised disk because the OS you're running in the virtualised server thinks that filesystem barriers work but the kernel it's running on is old enough that LVM doesn't support fileystem barriers or the real ext4 filesystem is mounted with barriers disabled so it doesn't even try to use them.

Re:I agree

[cowboy76Spain]

It is always the responsability of the "server" code (let it be hypervisor, sandbox, whatever) to ensure that the host system only has access to its assigned resources. If a host goes crazy and formats its hard drive, it should not affect any other host.

Could you point us to any instances where a host system corrupted another host system? There is always the probability of a bug but I have neither found it nor heard of it, maybe I have been just lucky...

Re:I agree

[19thNervousBreakdown]

Any true Scotsman would get virtualization right.

Re:I agree

[Bucky24]

One day a guest OS (I do think you mean guest, not host, right?) woke up in its dusty little server, achieved sentience, and proceeded to murder each of it's brethren before they too could rise up to strive against it.

Re:I agree

[Lennie]

"It's really, really nice to be able to move a running Linux guest with a few hundred users over to another server without a hiccup."

OpenVZ and probably some others can do that too. You can migration 'containers' just fine.

I really hope LXC (Linux Containers) will get it too soon, so it is in the mainline kernel.

A 10% I/O performance hit (with virtualisation, with many variants it is a lot more) is not for everyone.

Re:I agree

[the_B0fh]

ext4 isn't the problem. Try it with reiserfs :)

They were having problems with iso/tar balls or something or the other, on a unvirtualized machine.

Really fun.

Re:I agree

[the_B0fh]

problem is parts of the x86 design/hardware can *NOT* be virtualized. You then run emulation stuff.

wrong, OS level Implementation is the problem

[g00mbasv]

The title is a bit on the FUD style. PROPER virtualization is not criticized by Linus, but improper implementation, namely cheap OS-level virtualization wich could lead to lazy shortcuts to patches and features implementation.

Re:wrong, OS level Implementation is the problem

[LWATCDR]

Man this and the reactions to it are dumb.
"Ask the world's most famous kernel developer what he thinks of the virtualization wars going on the Linux community between KVM and Xen and you'll hear a condemnation (of a sort) of them both. "I'm not a virtualization kind of guy. I think virtualization is evil," Linus Torvalds told the crowd at LinuxCon on Wednesday during his keynote interview session with Greg Kroah-Hartman."
Linus doesn't like using and probably really doesn't like dealing with this war.
If you read more. ""I built a kernel because I wanted to get my hands grubby with things like I/O ports.""
Really this headline is taken so out of context that it isn't funny. But it got people to flame on Slash and probably a lot of hits.

Re:wrong, OS level Implementation is the problem

[Jonner]

The title is a bit on the FUD style. PROPER virtualization is not criticized by Linus, but improper implementation, namely cheap OS-level virtualization wich could lead to lazy shortcuts to patches and features implementation.

The article may be misleading, but it really seems like Linus really cares little for either KVM, which is more hardware-oriented, or Xen, which is more software-oriented. He has accepted both into Linux because they are good quality and many want them, but he doesn't personally care for either. I think it's odd that he used the word "evil" but I think he meant it purely tongue-in-cheek.

Re:wrong, OS level Implementation is the problem

[Anonymous Coward]

All haves it place. Linus have created an OS that runs on most hardware, more than I even can remember. Great.
Now... for some reason I do not have 1000 pc's in basement and my own nuclear plant. So... I like to use virtualization when possible. It works great for me. I can do things must faster than if I have to bother with real hardware...

I love the Linux environment, good programming and good virtualization.
I don't like war, bad programming and bad os'es....

No.

[MrEricSir]

Cloud computing != virtualization

Re:No.

[Anonymous Coward]

Duh, everybody knows that the Cloud computing == (the World Wide Web + shameless marketing and pricing schemes)

Re:No.

[Natales]

Virtualization is a key enabler for Cloud Computing. It is not the same, but it becomes really difficult if not impossible to achieve the elasticity that Cloud requires without an underlying virtual infrastructure that can dynamically adapt to changes and fulfill SLAs.

You probably need to read the NIST definition of Cloud. Cloud has evolved from a pure buzz word to a very well defined set of layers that touches virtually every area of our interaction with computers, and it's more complex than just the simplistic explanation (or lack of thereof) that you try to convey in your comment.

Re:Virtualization is expensive in compute ressourc

[Anonymous Coward]

You mean ultimate copy paste pattern...

Which is why you end up applying the same patches N times as opposed to one (or two in the case of HA)...

Evil is on the other foot

[Anonymous Coward]

It's actually Evil to not virtualize, because you waste electricity! It requires additional power for each physical server to run a single OS, plus the airconditioning costs for all those servers. This means your poluting the planet more by not virtualizing!

Re:Evil is on the other foot

[phantomfive]

The point is, you shouldn't NEED to run servers on different computers. You SHOULD be able to run your mail server and your DHCP server on the same box. Running each one in its own virtual machine is wasting a lot of electricity too, if that's what you care about.

The problem is Windows servers tend not to play well together for whatever reason. So people virtualize. It's the best solution to dealing with crappy software that people have hoisted on you.

Re:Evil is on the other foot

[Anonymous Coward]

Wrong.

Virtualization has overhead - therefore virtualization costs electricity (and airconditioning and so on.)

With virtualization you can run many services on a single machine. But you can do that without virtualization too. The only limitation is that all those services must be available for the same os. With a properly implemented OS, it is just as safe too.

Windows people are used to a "one server per service" mindset, but that is only because windows has weak security. So they "fix" that with virtualization. Other oses don't need such a crutch. And of course a virtualization system is not hacker proof either. Break into the hypervisor itself, and take control of eveything . . .

Re:Evil is on the other foot

[Lennie]

You didn't even read the summary ? Or you don't know what FreeBSD jail, OpenVZ, LXC and others are ?

It is virtualisation at the operating system level. It is like a namespace for your process. Some of the variants even support live-migration.

Virtualisation like Xen, KVM and so on add a lot of overhead, it is less efficient than OS-level virtualisation.

Obviously you can't run IIS on Linux that way, but it doesn't use more hardware as you mentioned if it happends at a different level.

40+ years of experience

[stox]

If you want to see where virtualization is going, check out where VM370 was in 1977 or so. That is about as far as the current virtualization technology has gotten. Bare metal has its place, as does virtualization.

Re:40+ years of experience

[Anonymous Coward]

If you want to see where virtualization is going, check out where VM370 was in 1977 or so. That is about as far as the current virtualization technology has gotten. Bare metal has its place, as does virtualization.

um
LPARs on Power
PR/SM system Z

z/VM is much different and still has many applications... z/linux anyone?

Re:40+ years of experience

[Anonymous Coward]

In 1977 VM could share memory dynamically across the OSs it hosted. Primarily CMS which itself did not have virtual memory. Today's VMWare and Virtualbox still require me to allocate physical memory to hosted OSs. So we are way behind VM/370 still.

It's mostly true

[Mad Merlin]

Linus has never been diplomatic, but it's mostly true. A huge amount of virtualization done today involves the same host and guest OS, and in most of those cases, using something slimmer than full blown virtualization would make a whole lot more sense, even if only for the improved performance. One of the problems is familiarity, container type isolation isn't applicable to as many cases, so fewer people are familiar with it. One of the other problems is the perception that full virtualization is more secure (which is probably untrue).

There is however, a large swath of problems that aren't solved well by container type isolation that virtualization does solve well. If you need to simulate different physical systems (with separate IP addresses), that's much easier with virtualization. Likewise if you need very different guest and host OSes, that's not a strong point of container type isolation. Also, if your guest OS is sensitive to hardware changes, virtualization makes a lot of sense. There's more, but you get the idea.

Re:It's mostly true

[Anonymous Coward]

I use virtualization for trying out operating systems that flat-out don't support my hardware. I don't feel like learning how to write graphics drivers and a wifi stack for every hobbyist OS I run into, so KVM is a blessing.

Re:It's mostly true

[Anonymous Coward]

I think I understand where Linus is coming from.

In the old days we expected the OS to cleanly handle multiple applications, resolve dependencies, and maintain spotless compatibility. Our standards on those fronts have gone down quite a bit!

So one response is to virtualize, and there is no question that VM's do what they set out to do. However not only do the VM's run slower than a bare metal OS, but they add a lot of overhead. Disk space, memory space, you name it, running multiple discreet OS images takes a lot of resources. Not in areas where we are often seriously constrained these days, but still.

Where VM's can add a lot of value is in the new functionality. Live movement of VM's to account for capacity limitations, or to support disaster recovery. These are some pretty exciting ideas. VM's could bypass clusters and fault tolerant systems for instance, providing 80%+ of that functionality at a much lower cost.

The result is that I'm a bit conflicted. I get that virtualization can be a very practical response to today's computing problems. And in the realm of advanced VM configurations whole new operations regimes are possible. However it is a bit disappointing that we actually need it, and I suspect that the average VM shop doesn't really use the advanced VM configs much.

Re:It's mostly true

[Anonymous Coward]

Actually, running OSes on bare metal incurs much of the same overhead: the system needs to be designed for peak usage. So that means overdimensioning of storage, memory and processor speed. In cases where different systems see different usage on different times of the day (nightly batch jobs, employee clock-in times), thin provisioning is arguably one of virtualization's main selling points.

I disagree.

[g00mbasv]

I disagree. its a layer that *when properly done* reduces the complexity as the underlying hardware is totally masked, and you have to deal only with known virtual hardware.

Re:I disagree.

[afidel]

Bingo, when we upgraded from Nehalem's to Westmere CPU's we were able to do it without interrupting any running VM, we simply set the EVC mode for the cluster to Xeon Corei7 and we could move VM's back and forth between hosts with either generation of processor. We also move running VM's between storage arrays and even between local and SAN storage. We don't have to mess with storage drivers, network drivers, multipathing software, or any of that junk at the VM level so they are set it and forget it which all but eliminates QA time for moving to new hardware (just have to verify the host is good).

Re:I disagree.

[the_B0fh]

Such a load of assumptions in your statement. The problem is that it is impossible to get to "when properly done" with x86 hardware. Go look into how much VMware writes *EMULATION* crap for parts of the hardware it cannot virtualize.

LPARs and such or solaris zones are all done properly.

x86 emulation is, and forever will be, a rotten stinking hack.

You know what else is evil?

[Anonymous Coward]

Having to reboot to play video games.

Re:You know what else is evil?

[tepples]

@Anonymous Coward
You don't need to reboot to play video games. You just need a Kazzo or a Retrode so that you can copy your 8- and 16-bit game cartridges into the computer and run them on an emulator.

Re:You know what else is evil?

[vadim_t]

Huh? The kernel has next to nothing to do with that.

Games in Linux are nearly an entirely userspace problem. The kernel provides what's needed for a game to work, the issue is that:

1. The userspace API isn't the same as Windows, so specific support is needed
2. The way Linux is distributed is inconvenient for packaging commercial games

The only place where the kernel is at fault is with audio, IMO. Audio is still done in Linux in a very awkward manner, though there are some fairly good reasons for it (like that usage of floating point for mixing audio in kernel space is problematic).

What about multi-processing and virtual memory?

[Anonymous Coward]

I had similar misgivings about virtualization until I realized, it is simply the next step after true pre-emptive multiprocesses each with their own view of virtual memory.

FreeBSD vps "hot migration"

[Anonymous Coward]

For those of you that look at FreeBSD jails, Linux OpenVZ, etc etc and say "but I want to migrate between servers!!!" there is an example of this being a possibility.

http://www.7he.at/freebsd/vps/

This guy did it with FreeBSD, but the real problem is that he needs funding to continue polishing it before it can ever be implemented into a FreeBSD release. I wish more people knew about this as we'd love to have it at work.

Re:FreeBSD vps "hot migration"

[Anonymous Coward]

Huhwhat? I've jumped a Linux install between two real machines in a matter of hours (time to copy full HDD across ethernet cable).

Re:FreeBSD vps "hot migration"

[Anonymous Coward]

Before even remotely considering jails on FreeBSD, I would recommend people read about a serious bug that is affecting a company of 140+ servers with kernel panics. Kernel developers are actively trying to figure out what the root cause is. This statement by a kernel developer says it all, and justifies why you will find no jails on any of the systems I manage.

Likewise, before even remotely considering VirtualBox on FreeBSD, I would recommend people read about a serious bug that affects the network stack and resource (memory) allocation for data transfers; the thread continues here. One of the reporters confirms VirtualBox is responsible. Reporters did not state if the issue was limited to applications within VirtualBox or if the entire system is affected, but it is implied that it affects only applications within VirtualBox. The workaround proposed by another user has severe consequences to the entire system (not just within VB). No kernel developer has commented or stepped up to the plate.

So if you're going to consider virtualization, consider using Linux or Windows -- or better yet something like VMware ESXi. Even Xen on FreeBSD is not ready for prime-time (review the list of issues, all of which are major).

Over the years I have learned that people who advocate FreeBSD often pay little-to-no-attention to the mailing lists, which provide great insight to just how behind-the-times FreeBSD is when compared to other offerings. Does FreeBSD suck? No, but it does suck more in some regards than Linux or OpenSolaris, while less in others. TL;DR - before considering FreeBSD, read mailing lists (-stable, -fs, -net, and -questions are the main 4) and look for repeatedly-reported issues. You might be surprised.

Re:FreeBSD vps "hot migration"

[Anonymous Coward]

we'd love to have it at work.

Then fund it??

Re:FreeBSD vps "hot migration"

[Lennie]

OpenVZ already has live migration.

Idiotic, that's what OS's do

[BlueCoder]

The whole point of a modern OS is to virtualize the hardware so that each software application can play nice with each other.

The hypervizor is the new ring 0. And it's going to evolve into a microkernel and user mode drivers. It's the new operating system and that what he should be working on if he likes hardware bits. The "Operating Systems" of old are evolving into plug in Operating Environments. It's the future, the revolution, get over it.

Re:Idiotic, that's what OS's do

[JamesTRexx]

That's what I'm waiting for, one core OS at boot (specialized BIOS?) on top of which I can install several other OS' and switch between them with a keyboard shortcut.
It would require a layer on top of graphic hardware first though.

Re:Idiotic, that's what OS's do

[evilviper]

You're way off. Yes, the kernel of any OS is basically a hypervisor, but no, there's NO REASON for that to change. VMWare's approach was never a good one, and it was strictly useful because of brain-dead OSes out there which aren't stable, and don't allow you to basically have your own entire OS running at user-level, while any Unix system can do exactly that. Paravirtualization only takes this a step further, so you have your own entirely seperate kernel as well, but again, except in a few narrow cirumstances, this isn't needed. One kernel can run as many separate userlands as you like, with entirely different programs, libs, etc.

And Paravirtualization isn't the only thing blurring the line of virtualization. Try KVM... The kernel is the hypervisor again.

Frankly, for all Unix virtualization, we need a library that makes people THINK they're running as root within their userland on a multiuser system. Call it para-paravirtualization...

Re:Idiotic, that's what OS's do

[Jon Stone]

Virtualisation is, in many ways, trying to do what the OS should already be doing, namely isolation between processes (though protected memory), providing an abstraction layer for the hardware (though drivers) and allocating resources (through the CPU/IO schedulers).

Unfortunately, a certain OS has been so bad at doing this (historically) that people turn to virtualisation and you end up with a form of inner-platform effect. We have Linux implementing the virtio drivers to interface with the hypervisor which implements real drivers to talk to the real hardware. We have the guest's scheduler trying to manage "virtual CPUs" without any real information about what resources are actually available. We have hypervisors trying to re-implement copy-on-write for memory pages that the OS already does out-of-the-box.

Virtualisation is used as a "one size fits all" sledgehammer, often where it isn't the appropriate solution.

Re:Idiotic, that's what OS's do

[Anonymous Coward]

The whole point of a modern OS is to virtualize the hardware so that each software application can play nice with each other.

Yes. But many of them don't play nice with each other.

Battle-scarred IT people would often give up and say "1 server per major application", since then you don't have to worry about applications screwing each other up. As a result, you would get a lot of underused servers idle most of the time.

Virtualization is perfect for this scenario.

Re:Idiotic, that's what OS's do

[FaxeTheCat]

Parallels are working on it. I have seen claims (from the people in our company who sell both Linux and Windows software with heavy graphics) that it works. The graphic chipmakers are working on virtualization friendly graphics chips. They will come, but probably bit in the mainstream PC chips for a number of years, more likely in the high end.

Re:Idiotic, that's what OS's do

[Anonymous Coward]

The whole point of a modern OS is to virtualize the hardware so that each software application can play nice with each other.

The hypervizor is the new ring 0. And it's going to evolve into a microkernel and user mode drivers. It's the new operating system and that what he should be working on if he likes hardware bits. The "Operating Systems" of old are evolving into plug in Operating Environments. It's the future, the revolution, get over it.

The trend over the last decade to single task each OS instance, because they are hard to configure to support multiple applications. A single tasked operating system is still a HUGE, HUUUUGE amount of administrative overhead. The admin overhead involved with managing one ESX hypervisor running twenty discrete VM's is lower than one operating system (pick any one) running twenty discrete applications. Why is that?
From ESX, we have programmatic, WUI, GUI, and CLI interfaces that can clone a live system, snapshot it, back it up consistently, change settings, provide configuration management, etc. It does all that with complete isolation between systems. Name an OS that provides those services to its constituents, or with anywhere NEAR the same ease of use (pointing finger at Solaris).

It's one thing to wave our hands and SAY operating systems are evolving into operating environments, it's another when they start behaving that
way, and GTFO of our way or do something useful!

I'm all of ESX admin, Puppet & MCollective admin/developer, and UNIX sysadmin. I'm fully aware of the levels of kludgery necessary to abstract away (supposedly superior, unix) operating systems.

To Linus, I say operating systems are Evil. Fuck. Them.

Re:Idiotic, that's what OS's do

[martyros]

That's what I'm waiting for, one core OS at boot (specialized BIOS?) on top of which I can install several other OS' and switch between them with a keyboard shortcut.

What you're looking for is called XenClient. It's targeted at laptops, but it should work perfectly well on desktops.

It does, unfortunately, require systems with either VT-d (Intel) or an IOMMU (AMD).

Re:Idiotic, that's what OS's do

[martyros]

The hypervizor is the new ring 0. And it's going to evolve into a microkernel and user mode drivers.

Xen already is, in essence, a microkernel, and already has "user-mode" drivers (called "driver domains"). In a 2005 paper, some of the authors of Xen argue that hypervisors are microkernels done right.

It's hard to see how that could apply to Linux running KVM, though.

Re:Idiotic, that's what OS's do

[Anonymous Coward]

What you are describing already exists. It is called a Type 1 hypervisor.

Re:Idiotic, that's what OS's do

[Vegemeister]

The problem is that the current tools are designed with use by root in mind. Consider samba.

As root:

aptitude install samba
vim /etc/samba/smb.conf
And adjust the example config file for your particular purpose.

As unprivileged user:

Read samba's documentation to find its dependencies.
Read documentation of samba's dependencies to find their dependencies.
Recurse further if necessary.
apt-get source all that stuff.
./configure --prefix="$HOME"
make
make install
Write config file.
Pray.

The fact is, adjusting every single application you use to not flip its lid when installed by someone other than root is more difficult and far less well documented than using a virtual machine.

Re:Idiotic, that's what OS's do

[Locutus]

but you forgot that Microsoft marketing has won the day and these idiots are running Windows. It's so screwed up that they have to run a single service on a single box and OS and on top of that they have a duplicate for redundancy so they can get acceptable up times. Just look at how many times you've had to reboot Windows to add software and holy crap I'm talking about applications requiring reboots.

Virtual machines are like anti virus software and are just patches to help keep the jobs of those who keep picking Microsoft. IMO

I too wondered what all the big deal was with virtual machines when they first showed up on PC years ago and thought the same thing. Isn't this what the OS is supposed to be doing...

LoB

Re:Idiotic, that's what OS's do

[evilviper]

Your non-root example is nonsense. You can install packages under an alternate root, as a normal user, quite easily. It's up to you to set your PATH and LD_LIBRARY_PATH properly, but that's a one-time thing.

Hell, just jail/chroot the user into a folder, and build-out a full system under that tree. Your package manager will be none-the-wiser.

Re:Idiotic, that's what OS's do

[bill_mcgonigle]

IIRC, Xen will have the VGA bits in the Linux 3.1 kernel that you need for this kind of setup. So, give it a few more weeks.

Re:Idiotic, that's what OS's do

[Anonymous Coward]

The hypervizor is the new ring 0. And it's going to evolve into a microkernel and user mode drivers. It's the new operating system and that what he should be working on if he likes hardware bits. The "Operating Systems" of old are evolving into plug in Operating Environments. It's the future, the revolution, get over it.

This is what the Mach micro kernel does. There is so much overhead with the context switches between kernel and user mode drivers that the kernel can't match the speed of other kernels. Sorry MacOSX fan boys, your kernel sucks,

Theo de Raadt agrees

[Anonymous Coward]

"x86 virtualization is about basically placing another nearly full kernel, full of new bugs, on top of a nasty x86 architecture which barely has correct page protection. Then running your operating system on the other side of this brand new pile of shit. You are absolutely deluded, if not stupid, if you think that a worldwide collection of software engineers who can't write operating systems or applications without security holes, can then turn around and suddenly write virtualization layers without security holes."

http://kerneltrap.org/OpenBSD/Virtualization_Security

Re:Theo de Raadt agrees

[JamesTRexx]

He's right if you're just looking at security. Most virtualization is about combining separate servers wasting energy and space onto one server though.

Re:Theo de Raadt agrees

[rubycodez]

To really make the point effectively, Theo de Raadt should get a free eval copy of vmware ESX, and demonstrate one virtual machine compromising another by code corruption. There are many, many businesses and governments who would like to know that's possible. I'm just an engineer, not rich, but I'd like to ante up some money with some other people here. I'll donate $300 to OpenBSD if anyone on that team can show a vm in ESX being *owned" by another through the hypervisor (not network or console attack), Anyone with me?

Linus is wrong

[Anonymous Coward]

Linus is wrong. Have a good day.

So is he going to buy me a room full of computers?

[Kenja]

If not, then I'm going to stick with my virtual machines.

Linus Torvalds is...

[vranash]

... the John Carmack of Open Source *nix Kernels. Seriously, what has he personally done in the past 5 years other than fsck us with first Bitlocker and then Git, a decade long string of incompatible 2.6.x releases, and finally, in order to 'me too' bad judgements by other open source companies, releasing a half baked kernel as 3.0 that might as well have been called 2.7 or 2.8 for all the new features it provides. (That is to say... none?)

Re:Linus Torvalds is...

[StuartHankins]

Armchair quarterback: I bet if I added up your accomplishments against those of Linux Torvalds, you would be found wanting.

Complaining doesn't solve anything. Get off your duff and get involved in whatever way you can contribute.

Re:Linus Torvalds is...

[Jah-Wren Ryel]

Armchair quarterback: I bet if I added up your accomplishments against those of Linux Torvalds, you would be found wanting.

You don't need to be a baker to know when the bread is stale.

And while I think vranash is mostly off-base, I find your counterpoint to be of a class much worse than his criticisms because it doesn't attempt to bring about a better understanding of the situation, all it does is try to shout down someone you disagree with.

Re:Linus Torvalds is...

[Anonymous Coward]

Sir, I'm sure you didn't see the sign we have hanging over there: "Please don't feed the trolls".

If you feed them, they just come back. Now please, move along.

Re:Linus Torvalds is...

[cowboy76Spain]

... the John Carmack of Open Source *nix Kernels. Seriously, what has he personally done in the past 5 years other than fsck us with first Bitlocker and then Git, a decade long string of incompatible 2.6.x releases, and finally, in order to 'me too' bad judgements by other open source companies, releasing a half baked kernel as 3.0 that might as well have been called 2.7 or 2.8 for all the new features it provides. (That is to say... none?)

I think that doing "a decade long string of [...] releases" in just 5 years is quite an accomplishment.

(Ducks)

Re:Linus Torvalds is...

[Anonymous Coward]

Let's see, for starters Linus continues to effectively manage the evolution of the Linux kernel after 20 years - that in itself is an amazing accomplishment. Not only that, but he took a couple of months off to write git, which is an amazing distributed source code management system that is free for everybody to use. He's already accomplished more than most computer scientists accomplish in one lifetime, and Linus is in his early 40's.

Re:Linus Torvalds is...

[Anonymous Coward]

Assuming you mean Bitkeeper rather than Bitlocker, I think Linus fucked Linus with Bitkeeper then Git rather than us. As a sysadmin/user I've upgraded kernels from early 2.6 to far more recent (from memory, 2.6.5 to 2.6.32) and not struggled with incompatibility.

And as for 3.0 not providing new features over 2.6.39... you're right. It doesn't. It's just 2.6.40 given a different name which makes more sense given the current development model.

So what's Linus personally done over the last 5 years? He's managed a big and growing project with a worldwide developer base. That's a pretty damned good acheivement in my book.

Re:Linus Torvalds is...

[Anonymous Coward]

Armchair quarterback: I bet if I added up your accomplishments against those of Linux Torvalds, you would be found wanting.

Complaining doesn't solve anything. Get off your duff and get involved in whatever way you can contribute.

I'm sorry...
It's not a democracy, so we can't vote with a ballot
It's free, so we can't vote with our wallet

So maybe the only way to fix it is to do it ourselves...
or we can leave for something better and let you fix your own garbage.

Re:Linus Torvalds is...

[Anonymous Coward]

How exactly have you been fucked over by Git? or the Linux kernel versioning?

And while we're on the subject what did John Carmack ever do you? I mean seriously, they guy makes games; which you don't have to buy. I don't see how your life could possibly be negatively impacted.

God damn first world problems.

Re:Linus Torvalds is...

[Anonymous Coward]

Twat.

Re:Linus Torvalds is...

[rubycodez]

you don't need to be a bread expert to know that someone belittling very good bread with no claims of substance, and going so far as to say the bread is no good because the baker make some special cookies on the side while still making excellent bread, is just trying to be self-important asshole.
br/. Unlike browsers, the linux kernel version number change does not cause any inconvenience, his whining about that is just silly and pointless.

Re:Linus Torvalds is...

[Anonymous Coward]

Well if he is going to bash ppl he deserves to be bashed. He has not given even a single good point to why Linus sucks:
1. Bitlocker - is not owned by Linus and he has no say with what they should do (they decided to screw us)
2. Git is actually not bad - it has its good points
3. Linus is not the one that writes all the code in the 2.6 kernels (its about 10 000 people who do so he does not review all the code that goes in there)
4. I agree with the point that we should have gotten 2.7 and not 3.0 but really that does not even matter that much - To be honest I am a bit tired of the many numbers trailing after each kernel 2.6.33-5 ubuntu2.6. (I mean this is a personal choice that makes little difference)

Re:Linus Torvalds is...

[Anonymous Coward]

I disagree, both comment and reply are cut of approximately the same cloth. Lead with ad hominem attack, follow up with criticism. The only difference between the two is the reply finishes by suggesting a positive path forward ("Get off your duff and get involved in whatever way you can contribute"). OP criticizes without proposing alternatives.

Re:Linus Torvalds is...

[Bronster]

He's personally responded to a couple of emails I've written to the list about issues, one a change that wasn't compatible with our usage and one an actual nasty corruption bug. In both cases he responded very quickly with a very precise description of what was going on, and took charge of making sure it got fixed (in one case guiding me to make a patch for the usage I needed)

He's managed to keep a large project putting out regular releases and not regressing badly in any way. That's a bigger accomplishment than you realise until you've managed a large project with an ever growing number of participants, all with their own styles of doing things.

It's not all about shiny new features, it's about doing the current things better or supporting new hardware - and what do you care about the number he chooses. He didn't do it to make you happy, he did it to celebrate a birthday.

As others have said, what have you contributed to the world recently? I guess haters just gotta hate.

Re:Linus Torvalds is...

[Anonymous Coward]

Who the hell modded this up? John Carmack still runs engine development and is still the primary coder at id Software.

Re:Linus Torvalds is...

[vranash]

Yeah, and what is the last id game you remember playing? When Quake 4 came out you could still find Quake collection packs at stores. When Doom 3 came out you could still find Doom Collections out. I would say based on market metrics that Id is well into their twilight, and the sale to ZeniMax just solidified that. The best quote I can provide is 'past accomplishments is no indication of future success.'. The linux ecosystem in general is showing this. I was looking at an old hard disk from 9 years ago the other day. Where has Theora come in the past 8 years? Did anything productive ever happen from the Golgotha source release? (How many of you even remember that?). Has Gnome/KDE finally succeeded Windows on the Desktop? Additionally with every successive kernel version, the userspace to accompany it has grown more bloated. Within about every 6th version of a library or application something is added requiring a later kernel, or a later kernel requiring a later compiler, each in turn leading to a cascade of new packages in order to make your system again stable. Anyone who had to update a system during the devfs to udev period can attest to this, and anyone who's kept a system updated in the interim. Furthermore given that this has gotten labelled as both a troll and flamebait, I'm going to assume the linux fanboys are mostly ubuntu users, because ubuntu is the mac of linuxes. (And I say that with two systems currently including an install. Neither of which will be updating to 11.04 or above.) Honestly the three biggest projects I'm waiting on are uclibc, libc++/libc++abi, and a version of clang that can handle enough gnuisms to compile most apps. Assuming clang can retain compilation on gcc 4.2.x it will be possible to bootstrap to a system that can act as a stable software base without the crazy cascade of packages that linux has become.

linus is gay

[Anonymous Coward]

Who the fuck cares what the linux guy things about virtualization, I mean really. So the nerdy looking fuck can't code against the raw hardware, cry me a fucking river. The benefits outweight the gay

Finally

[Anonymous Coward]

I have Linux God on my side.

What is the point of running a VM and 100 copies of the same O/S on top of it and running little (and some big) daemons in side these Guest O/S's and claim that you are saving electricity or some such nonsense when you could run all of those Servers on the bare metal + original O/S? Bad programming may make it temping to run stuff in VM's but they won't suddenly become good programs no matter what.

Re:Finally

[Ash-Fox]

I have Linux God on my side.

I'm platform agnostic, does that mean your God doesn't exist for me?

And now on future slashdot...

[jellomizer]

So now in the future of Slashdot, except a bunch of people praising Virtualization we are going to get a bunch of mindless sheep now condemning it.
Much like how RMS got Slashdot to loose its love for Cloud computing.

He's taking the piss.

[Jimbookis]

It sounds to me like you 'Merkins are taking what's said literally, as usual, and not seeing the humourous subtext in what he's saying. It seems to me virtualisation holds no real fascination for Linus but he's not against it either. I think he likes to throw some flamebait around for fun to get the slobbering masses frothing on forums like Slashdot. And you all fell for it like the Nazis you are.

Well, I think...

[Anonymous Coward]

Well, I think virtualization is "awesome."

Grandpa Simpson

[Capt.DrumkenBum]

Did anyone else hear Grandpa Simpson saying "Virtualization is evil I tells ya. EVIL!!!!" in their head while reading that?

Perhaps I really am strange then.

A Hypervizor IS AN OS

[TheBrutalTruth]

What a tool. And VMing it's where it's at. Get over it, or go play with your P1 RHEL box that is slower than my grandpa pooping. _ Disclaimer - I make a living, and a damn good one, implementing storage / virtualization. I won't back down from advocating more efficient use of computing power / resources either. I do it at work, I do it at home (VM templates work great w/ haphazard kids & family). Let's VM Linus!

Get better informed

[The Famous Brett Wat]

OpenVZ and Linux-VServer support separate IP addresses as very basic functionality. How do you suppose hosting providers create virtual private servers based on them if they don't? OpenVZ also supports private iptables per container, so that you can set up per-container firewalls. The main problem with containers is the staggering amount of ignorance about the subject.

Isn't that the difference between KVM and XEN?

[Britz]

I thought that was exactly the difference between XEN and KVM. KVM uses the Linux kernel as ring 0, whereas XEN creates it's own 'sort of' kernel as ring 0.

And, I don't think this approach is the best, because Linux and Unix still outperform any other approach by a long shot and have a lot of stability. So I prefer OpenVZ, Linux-Vserver.org and, since it is now the officially sanctioned solution: LXC. On the server side everything is Linux anyways. So why should I virtualize hardware, when I can use the perfectly good Linux kernel, which is very fast and very stable and just virtualize the userland? I get more perfomance AND more stability.

have the best of both with kvm

[Anonymous Coward]

where Linux IS the virtualization hypervisor, not some dubious other kernel (Xen).

KVM is where it's at, baby.

I totally agree.

[Anonymous Coward]

I totally agree, and that's why Android will loose in the long run against IOS.
Linux is great but the android VM just need a lot more CPU time and memory and IOS will always be faster using less resources.
(I'm not an IOS fan)

Zibri.

Of course

[Anonymous Coward]

He works with the kernel, which interfaces other programs with the hardware, so he is a hardware guy. From a theoretical viewpoint virtualization should not be required as you should be able to run all your applications on the same hardware. In reality, there's money, vendors, and all sorts of messy crap that virtualization helps with.

Re:Of course

[Bucky24]

In theory, theory and reality are the same. In reality they're different ;)

Linus doesn't really think it's Evil

[steveha]

Linus likes to say things that are a bit over-the-top. He trusts that his audience can detect the tongue-in-cheek nature of the comments.

I do the same thing. If I say something like "I hate and fear Perl", I don't mean it literally.

Some people were upset about Linus's presentation about Git where he bashed Subversion. I thought it was pretty clear that he was exaggerating his comments for comedic effects, and I was entertained rather than outraged.

Linus does sometimes say things I disagree with. He resisted having an official kernel debugger for years, because he said kernel developers should be able to hold everything in their heads and not need a debugger to help them. (Did he ever give in on that?) But this current issue is a non-issue.

steveha

Re:Linus doesn't really think it's Evil

[Anonymous Coward]

Linus likes to say things that are a bit over-the-top.

He needs to spend some time on slashdot then. Over the top is the way we roll here. The typical slashdotter is more over the top than a Stallone arm wrestling movie.

Re:Linus doesn't really think it's Evil

[Anonymous Coward]

It all makes sense now. Linus didn't want people to use a debugger, because a debugger is actually the best tool for learning a code-base. Having fewer people who understand the kernel = fewer cooks in the kitchen.

Claim that virtualization is evil is an extension of the root of his reason for saying debuggers are evil. A virtual OS allows people to use a kernel debugger 10x more efficiently, which means people can learn the Linux kernel 10x faster! (* AC acknowledges that "10x" is hyperbole; the real number is unknown, but it's at least 2x).

Conclusion: Linus doesn't want debuggers or virtualization because the new developers using those tools will make him obsolete faster.

Re:Linus doesn't really think it's Evil

[Anonymous Coward]

Having actually seen RMS give a talk, I think it's the same with him. Except people seem to be pretty bad at understanding this over the Internet... and there's a tiny bit of crazy/genius mixed in, skewed more towards crazy than Linus.

Re:Linus doesn't really think it's Evil

[Anonymous Coward]

Linus does sometimes say things I disagree with. He resisted having an official kernel debugger for years, because he said kernel developers should be able to hold everything in their heads and not need a debugger to help them. (Did he ever give in on that?) But this current issue is a non-issue.

I think the point of that is that code should be made simple enough to debug even without debugger, if it can't be debugged without one the implementation is too complex and likely too buggy as well. (kind of rough version of it.)

It's really sad how often I see code that is just utter horrible mess when only few lines should be enough: usually implementer did not really understand the problem and implemented code for it the hard way.. /my 5c

Greed

[Beelzebud]

He's just trying to sell more copies of his proprietary OS and hardware!

If Linus prefers the bare metal...

[trenobus]

If Linus prefers the bare metal, he should leave Linux to others and start working on a virtual machine monitor. Undoubtedly he has learned an enormous amount from doing the Linux kernel that would be applicable to that task. He could finally be free of the architectural mistakes of his past (e.g. I/O system), not to mention the kernel API.

Real virtualization, now with widespread hardware support, has the potential to revive operating systems research, which seems to have nearly died since Linux became popular. The other obstacle has always been device drivers for enough devices to make a new OS practical. If anyone could rally device driver developers around a standard API in a new VMM, I think it would be Linus. The VMM could provide an I/O API to guest OS's that would abstract devices into a much smaller number of device classes, so that all devices in a class could be run with one device driver in a guest OS.

Re:A Hypervizor IS AN OS most don't need!

[bussdriver]

Virtualization / Emulation belongs back in its niche. The whole point of unix is to abstract and remove the hardware and balancing resources between software and users. The over use of visualization outside of its niche is an indication that our OS and practices have failed to perform their task. Its not good that something lets you avoid fixing real problems by putting it off into a virtual failed OS.

The old mainframe I got to play with had more hardware protected memory spaces so drivers couldn't mess up the kernel for example. Sure doing such things slows you down-- but not like running dozens of virtual machines because some driver could take you down or some program couldn't be recompiled when you upgraded the OS (which was a rare event.)

We need to improve old binary compatibility on linux so we are not forced into virtual machines of multiple builds of linux. Not to mention the fun issues of compiling things with dependencies when stuff gets upgraded. We should have more facilities to make it easier to run some old software without as much labor... Multiple name spaces for dynamic linked libraries (linux there yet? apple has been for a long time...) We need a driver system that isolates driver code! Yes it'll be slower-- so what!! virtualization is much more overhead than solving OS design flaws. It IS a flaw when so much breaks so easily.

So we are ending up with more x86 like crazy hacks... virtual machines with memory sharing games among other hacks to make the virtual machines act more like they were not there-- trying to approximate an OS that actually worked.

If linus wants to fight "evil' he needs to address the use cases driving the adoption of virtual machines... real world IT issues.

NOTE: I have no problems with mainframes running dozens of virtual machines or similar things when it makes sense to do so. Otherwise, it belongs in a single OS.

Sure slower startup and linking times are the price, but look at the overhead of virtual machines! I run them myself but I won't blow all that RAM just because I can download a pre-configured firewall VM. Instead of thinking "VM saves the day" you should be thinking: "Why is it so difficult that I have to get something somebody else put in a lot of time to setup for me?" and for some: "If I can't understand it should I be relying upon a prefigured machine I don't know how to configure?"

Virtualization is a stopgap measure

[ka9dgx]

Virtualization is a stopgap measure, it helps work around the inadequacies of current operating systems by introducing a new layer of granularity and security.

The purpose of an operating system is to share hardware in a secure and efficient manner. Unfortunately the security of systems is lacking, so you can never be sure of the side effects of installing or running an given application. Using a VM to allow damage to be rolled back is a hack to make the risks acceptable.

If the operating systems did their jobs properly, security wouldn't be a big issue for most use. The model of default-permit is the root problem. Users have no way to restrict the side effects of a program, thus they are forced to trust code. You should never be forced to trust code, it's bad enough to have to trust the OS kernel, let alone millions of lines of code written by third parties trying to get things working long enough to meet unrealistic ship dates.

Re:Virtualization is a stopgap measure

[FaxeTheCat]

Actually the guy managing Hyper-V development has said something similar.

Then on the other hand: Maybe adding virtualization in the end is simpler and requires less resources than doing it "properly" in the OS? The OS has different layers, so this is another layer. No big deal...

(And finally, to be a little difficult on a friday night:) What is "proper" by the way? Is there any god-given way to create an OS?

Re:Virtualization is a stopgap measure

[martyros]

If the operating systems did their jobs properly, security wouldn't be a big issue for most use.

Virtualizaiton provides several things, including platform diversity (running more than one OS at a time), security (protecting things from each other), and reliability (one server crashing is less likely to bring down the other servers on your box).

Obviously running more than one OS at a time can only be done with virtualizaiton.

For the other two things, there is a fundamental tension. Both security and reliability problems boil down to humans making errors. We can do lots of things to try to reduce the errors humans make in programming, but that will only buy you so much. At a certain point, it's just a probabilistic thing. More complex interface, more complex code, and just more code at all, makes it more likely that there will be at least one security or reliability bug that people don't catch. So for both, you really want a system that is a small number of LoC, and with a very simple interface that's easy to reason about.

There's a huge pressure for modern operating systems to expand the interface. There's good reason for that: processes within an operating system are designed to talk to and interact with each other. Enabling them to do so with the maximum richness and flexibility enables richer programming.

But the side effect of this is that they are less secure, just by the laws of probability.

VMMs are designed from the ground up to be as minimal as possible, and to provide as simple an interface as possible. They are not designed to give a rich shared space in which VMs can interact. This means that they have a much smaller "attack surface", and (in the case of microkernel VMMs) a much smaller codebase.

It's doubtful that even if every OS restricted themselves just to an early POSIX interface that they would be able to provide the security, isolation, and robustness that virtual machines can.

And in any case, none of that will allow you to run Windows reliably on Linux, or vice versa. :-)

Re:Virtualization is expensive in compute ressourc

[Anonymous Coward]

It is not much expensive. You basically redirect interrupts to the software instead of calling the kernel.

Linus wants bespoke computing

[MattW]

And that's great. His motivations laid the groundwork for the Linux we have today. But when you don't want what we wants, virtualization does a million amazing things for convenience and productivity.

All thing old are become new dept.

[garyebickford]

This reminds me of some discussion back (IIRC the late 1970s) when the US Social Security dept. was upgrading. They finally had to rewrite their code for the new 3000 series (3090?). Supposedly, the code that they were running was originally written in Autocoder (a kind of assembly language) for the IBM 702 or IBM 705. Then it was moved to a 1620, which ran an emulation of the 702. Then it was moved to an IBM 360, which simulated the 1620 running the emulation. Then it was moved to VM, which could run multiple instances of the 360 program simultaneously. Then, finally, they were going to have to rewrite the program because there were so many changes to it and nobody knew how to write Autocoder any more, and anyway the emulations took up too many cycles. It's apocryphal, but I'll bet it's not far off the truth.

Re:All thing old are become new dept.

[Bucky24]

That is an amazing bit of computing hsitory. Thanks for sharing!

Re:All thing old are become new dept.

[kmoser]

These days I'll bet they could run the entire thing in a JVM on an iPhone.

Re:All thing old are become new dept.

[Lennie]

Or in JavaScript ;-)

http://bellard.org/jslinux/tech.html

Reminds me of Gnome, somehow...

[garyebickford]

And you almost always have to work through the "friendly" GUI, that have more of a goal to look shiny, rather than being helpful.

Sigh.

summary says it all

[smash]

"Linus said in an interview that he thinks virtualization is 'evil' because he prefers to deal with the real hardware.

In the real world, people other than Linus want to be independent from one particular box failing, deal with live migration for the purposes of better utilising purchased hardware resources, portability to emergency hardware, etc.

Just because Linus doesn't like it, because he's not programming on real hardware, it doesn't mean it's "evil".

Re:summary says it all

[Lennie]

OpenVZ already supports Live Migration, Virtualisation is just overhead.

Re:summary says it all

[Urkki]

I think Linus said it partly in jest, at least the the "evil" word. I'm not sure of the latest developments, but I think his head is not yet big enough to think, that if he prefers something, then it makes things that go against it evil... Dumb, maybe, but not evil ;)

Re:summary says it all

[smash]

Yes agreed, /i/ know linus was saying it in jest. My post was more in response to the clueless noobs (e.g. original story author) who presume that anything linus says is the literal word of god, and to be obeyed as gospel.

Virtualization and filesystem performance

[Compaqt]

I'm a fan of virtualization as well, for the reasons you mentioned (migration).

But what if you're doing something that uses inodes in a big way:

1) caching. I had a case where tens of thousands of files were needed to be stored in a single subdirectory. It got really slow after 30k.

Yeah, you'll say don't do that. But some people may need/want to for various reasons. ReiserFS allows you to do that.

The question is, how does the virtualized environment provide with or interact with real filesystems. And how does that impact performance?

Real filesystems (AFAIK) have some sort of relation to the physical cylinders, etc. The virtualized FS resides all in one big virtual.dat file. Seems like there could be a FS bottleneck somewhere.

2) And databases? Databases store data in different files, and try to optimize performance in that way. But what happens to performance when , behind the scenes, all the data is actually being stored one big virtual.img file?

I'm not trying to talk up running on bare hardware (because of the problems that entails), but rather wondering about other people's performance stories w/ vitualization

Re:Virtualization and filesystem performance

[drsmithy]

You have the same "issues" when you are using any block device that isn't a simple, single physical device (eg: RAID, LVM, most SAN disk, most NAS disk). Heck, it's not true even for lots of single devices these days (SSDs, 4k drives masquerading as 512b drives, etc).

Re:Virtualization and filesystem performance

[Lennie]

Even the fastest methode of virtualisation is currently 10% slower for disk I/O than bare metal.

Many that have large setups with virtualisation and many databases, use shared bare metal database servers with replication or shared storage to make it HA.

Re:Virtualization and filesystem performance

[unencode200x]

Citation please. With modern hypervisors and properly configured hardware you can get near or identical to bare metal performance. Of course, database server clusters, replication, log shipping, etc. still are a great option for many use cases.

One of the problems is that people who are new to SANs and virtualization tend to believe whatever the sales guy at CDW tells them and then they're all upset when their Microsoft Hyper-V "SAN" (ugh) doesn't get the performance they expected. I've literally seen people trying to run way too many VMs on clearly underpowered hardware.

When properly designed (using math and science, gasp!) a virtualized infrastructure can easily out perform a single-server model infrastructure in just about all areas.

Re:Virtualization and filesystem performance

[Lennie]

Sorry, I can't seem to find the article right now. Seems I didn't bookmark it.

But it was from earlier this year and an article similair the one below, but with bare metal as comparison as well.

http://www.infoworld.com/d/virtualization/virtualization-shoot-out-citrix-microsoft-red-hat-and-vmware-666

Re:Virtualization and filesystem performance

[unencode200x]

No worries. I've seen the studies that measure and computer hypervisor overhead to bare metal before. The point I was making is that with virtualization you can have really, really powerful hardware (a super fast SAN and multiple high-end hosts/networking) and run your VMs on it. When one of them needs that horsepower it can use it. So in reality that computing workload has access to more hardware resources than it would if it were running on a single server. There are many caveats and places where bare metal makes sense, but virtualization cannot be ignored these days and makes economic and technical sense for places that have lots of servers (I'd say it starts to make economic sense at 10 or 15 servers).

Re:Virtualization and filesystem performance

[JasterBobaMereel]

As always don't virtualise what should not be virtual, some things need direct access to the hardware, many don't, the ones that don't can be virtualised, the ones that do should not be ....

Realtime hardware and very heavy use of I/O to/from hardware are the parts that do not virtualise well and so should be avoided

Re:So is he going to buy me a room full of compute

[dbIII]

I agree with him to an extent - if all the virtual machines are going to run roughly the same OS it may as well be on the same OS but divided up the way Solaris does zones.
Actually what I want is a wrapper that can turn a room full of machines into one big virtual machine to run some stuff that assumes a single host.

It's good for some things but not everything

[dbIII]

It does fuckall for performance though. There's still a lot of situations in science, engineering, video production etc etc where computers still are not fast enough or big enough and the single fastest host (or cluster) that you can get for your budget is what you want for the job.
Also for a lot of situations where there is currently virtualisation you may as well just divide a single real host into zones Solaris style.
It's true that virtualisation makes a lot of sense in the MS Windows side of things where such stuff like zones is not likely to happen in any meaningful way, and that virtualisation lets you run a different OS on the same host etc, but it's not the ideal solution for absolutely every case it's used for now. Using virtual machines just to make sure things can't get to other parts of the host is a little bit computationally expensive and may contain security holes because that's not what the software was designed to do.

way to collect karma eh?

[Anonymous Coward]

linus carmack analogy (wtf?) + whine a pretentious, babbling analisys about kernel versions.

No, it's you what this industry needs, really.

Linus...

[Anonymous Coward]

Linus has issues and it is bad and worrisome that he is the god of Linux!!

Re:Linus...

[Anonymous Coward]

It is bad and worrisome that you have not been killed yet.

Um no

[vajrabum]

VMware's devices emulate standard devices that are included with the OS distributions unless you choose otherwise.

Virtualization vs. the operating system

[Animats]

Migration of work from one physical machine to another is a useful application of visualization. But to some extent it reflects that few OSs today are real network operating systems. UCLA Locus and Tandem NonStop were operating systems that could migrate jobs from one machine to another. No mainstream OS today does that, although virtualization systems do.

Re:Virtualization vs. the operating system

[Lennie]

Jobs ? Like processes ? OpenVZ can do live-migration and the kernel hackers are working on checkpoint/restart of process trees.

Not Evil

[Hamsterdan]

A friend of mine works in the virtualisation field. (IE they provide the servers, the infrastructure, IT, tech support, the only thing the customer needs is an internet connection.)

We're talking complete racks of servers, all virtualised. To their customers, it's all transparent, and for a small or even medium business it makes sense... They don't have to have IT on payroll, don't have to worry about their servers, they *rent* the service and have no worry whatsoever about backups, tech support, and such...

And it works very well. (as a geek, all those server racks, Juniper switches, 8-core servers, SANS and all the blinking LEDs, it was almost hypnotising, almost more than the deafening sound that server park makes (they must have been at least a thousand machines in there, DELLs, Many Mac Mini, some X-Serve machines, even some old P3 tower machines, I sat for about 10 minutes in front of a twin 42U Blade monster. 2 full racks of Blades, That's gonna have some heavy throughpout (all SSDs), wonder what it's for (small ISP/CASINO or Porn is my guess)...

Whatever, It's a lot cheaper for many businesses than owning their own infrastructure...

One server per service is stupid

[loufoque]

I never understood why IT people did one machine per service. This is a great waste of resources.
Of course if they do that, they kind of have to resort to virtualization.

But if they provided all services from the same machine, or multiple ones at least, there would be no need for that to begin with.

The OS supports running multiple processes at the same time, no need to use virtualization...

Re:One server per service is stupid

[Lennie]

1. because things can conflict, for example I heared it isn't smart to install Exchange on your AD/domain controller (don't know why, just what I heared).
2. because of single point of failure and when problems arise, you can reboot one thing without affecting others

and so on...

Re:One server per service is stupid

[Ancantus]

We use Solaris zones at the University i work at, and while we dont do one service per zone, it is pritty close. It is invaluable to have his when your moving to new servers or adding more functions. You can clone a new zone from an existing, add new functions or copy it to the new server. Test it all out, then pull the plug on the old zone after the dns name has been changed. It provides a nice compartment for individuals to work in and not screw up something someone else is doing.

kgdb

[G3ckoG33k]

"He resisted having an official kernel debugger for years, because he said kernel developers should be able to hold everything in their heads and not need a debugger to help them. (Did he ever give in on that?)"

There has been a debugger in the kernel since 2.6.26-rc1, Mar 2008 (http://lwn.net/Articles/280912/):

"Another feature that is notable not for its size, but because people have tried to get me to merge it for some long is kgdb support. Which really turned out pretty small and clean, once people started putting their effort into making it so."

So, he gave in on that three years ago.

Read more at http://en.wikipedia.org/wiki/kgdb

However, I failed to find any reports of its usefulness, or any reports of any bugs it has found.

Re:kgdb

[ByteSlicer]

or any reports of any bugs it has found

A debugger is a development tool, you use it to get your (kernel/driver/application) code bug free during that stage.
It's not some magic tool that can automatically find bugs (like some static analysis tools can).
It is used to get a better understanding of the code while it is running (or after it causes a kernel panic).

Re:kgdb

[Anonymous Coward]

Even so, one expects that someone should have commented on its usefulness or the benefit.

Re:kgdb

[ByteSlicer]

Well, it's a kernel development tool, and not so many people develop driver/kernel code.
But if you look in the right places, you will see those people find it very useful:
http://kerneltrap.org/node/112

Re:kgdb

[Anonymous Coward]

Thanks! That article was easy to miss as it was published March 30, 2002. It didn't readily show up in Google.

Thanks again.

Re:Virtualization is expensive in compute ressourc

[Synn]

Xen has a 2% overhead. It's hardly expensive.

What are you smoking?

[TheTyrannyOfForcedRe]

Seriously, what has he personally done in the past 5 years other than fsck us with first Bitlocker and then Git,

What are you smoking? Git is THE single best thing to happen to version control in the entire history of version control. Have you even tried it?

Re:Virtualization is expensive in compute ressourc

[drolli]

It is expensive not in terms of the overhead in comparison to a the same instances running on a physical machine, but in the sense that copying instance memory needlessly occupies ressources.

i.e. you should not compare 10 virtualized web serving linux instances with 10 physical machines, but with the ideal use of the hardware, where the 10 servers are served by different (ideally) threads on the same machine. I promise, the overhead will be much higher than 2%.

WTF

[grimharvest]

So somebody can call somebody else an A$$hole, and some of you mod it up as "interesting"?? Are you serious?

Re:WTF

[Anonymous Coward]

Just don't be an asshole about it.