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

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

[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.

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.

No.

[MrEricSir]

Cloud computing != virtualization

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.

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: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.

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.

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

[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: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.

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:Some might argue

[19thNervousBreakdown]

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

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.

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:Some might argue

[Per Wigren]

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