FreeBSD's 7 arches probably cover 99% of the computers people would want to use.
Incidentely, of the 7 computers I have at home here, there are 5 where FreeBSD 7 does not run, and 4 on which no version of FreeBSD can boot even in theory.. Those are 2 HP/9000 systems, a Sun Sparcserver 20, an Amiga 3000, and a slightly older x86 machine. 2 of those are in use everyday (the sparcserver is internal dns and the older x86 machine is a slightly 'advanced' x terminal.
While NetBSD has sacrificed features and speed for portability, FreeBSD has managed most of the portability (from a practical standpoint)
No they have not. FreeBSD 7 does indeed run on many fairly modern machines, but does not run well on anything a few years old regardless of what cpu and architecture it uses.
What NetBSD's 'portability' results in, and which you seem to not realize at all, is in much cleaner solutions that will not break on some obscure, seemingly unrelated changes. You may not realize how important that is untill you tried to write an OS yourself maybe..
while adding new features.
Lets see... FreeBSD's usb support comes from NetBSD, so does its bluetooth support. Ah, but FreeBSD did add wireless and acpi support more decently during that same time..
Seems like a pretty even score to me actually, even while I am sure I forgot some things on both sides.
OpenBSD has a good niche, as security is a goal for while people are willing to sacrifice some features and speed. Portability alone is a strange goal however, since the only question that really matters is "does it run on all the computers I have".
"I want to use the same base system on all hardware that I have" is not strange at all. You obviously never have been in a large environment where different kinds of hardware are used because they are good at different kinds of things, but people would still want to run the same basic OS on them. Not everything is a PC, and that is good because PCs really suck at some things.
I'm a Linux user myself, so I don't have much reason for favoritism for a particular flavor of BSD. I try to keep up with OS news in general however......
I don't have anything against NetBSD, though I can't say I'd miss it that much (just like slackware as a Linux distro - historically important, but now largely irrelevant).
And that is why I will now say that you really have very little understanding of what NetBSD, FreeBSD, OpenBSD and the like are doing, about their similarities and differences, and what the relevance of NetBSD is as a system. It is sadly mostly irrelevant from a user point of view, it is definitely not from an OS developer point of view, and as such it has quite some relevance for even those who develop Linux right now.
If I had any current concern, it would be that I really hope DragonflyBSD succeeds, as it is pursuing some really interesting ideas in modern OS design.
If you'd have read the article and understood it somewhat, you'd realize that if anything, Dragonfly suffers from some of the exact same issues that resulted in the current state of NetBSD, and from my point of view, their relevance is similar, relevant to OS developers, not to end-users. That however does not mean that there is no reason for it to exist, and I fail to see any argument in your post suggesting why that is different for NetBSD.
I once by curiosity found this excellent article describing AS/400 pricing systems.
AS/400s are mini computers, not mainframes..
And I was shocked. If you buy the low-end version, you already have the full power chip, but just slowed down. For about $ 20k per processor you can unlock that, but that will also require you to buy another license, for several additional k$ per processor.
Usually that also comes with on-site service and the like of a kind that Dell and friends haven't heard of yet.
Also check the price/performance comparison for competing UNIX and Linux systems shown there. What kind of pricing system is that?
I don't know this particular study, but I'd mostly be looking for an AS/400 if I'm looking for something to run a huge database, not hundreds of virtual machines.
I can imagine the hardware is a bit more stable etc, but relative performance of the Power PC is not as good it used to be.
The performance of a machine depends on a lot more then the CPU however. IO bandwidth is a major factor in performance, esp. when the dataset you work on gets substantially bigger then fits in cache or even main memory.
Hence looking at the cpu architecture is not going to tell you much about a system which specializes in IO bandwidth (mainframe class computers)
Stability of hardware, possibility to replace hardware without taking the system down and such are also part of the picture of course, but not unique to mainframe hardware.
Still IBM is probably just asking prices that it thinks it can get away with. Why are they doing this?
Uh, because they are a business and are in it for the money?
And, wouldn't someone else be able to use commodity chips (some workstation-grade core 2 duo or opteron) and build a robust system out of that, for only part of they price they ask.
Seeing the amounts of money going around in that market, someone would have tried.. As a matter of fact, people tried more then once, and in the mid 90s the mainframe market looked dead and it looked like commodity hardware was outdating it. Somehow, this never really happened however.
Thomas J. Watson was a very good salesman (a bit too good to be legal:) ), and I guess IBM still has a very good salesforce to get their stuff out, but how long can they keep this up with pricing systems like this?
Untill someone beats them at it or the market just dissapears.. Right now however it looks like they beat everyone who tried to beat them.
Welcome to the real world where businesses exist to 'produce shareholder value', or more generally, are in it for the money.
Uhm, you're mixing timelines again. CAM wasn't until 1998 (or so) and replaced the old SCSI subsystem, ATAPICAM came after that again. And the ULE scheduler is important TODAY, while I'm talking about what was important in 1993-1995.
I remember buying my last scsi cd writer in late 1998 (a 4 speed plexwriter) which was actually slightly cheaper then an equivalent quality ide writer at the time. The same was true for the phillips cd writer I had before that.
It meant having to have a scsi controller, sure, but those didn't cost the world either, and untill the widespread use of 'burnfree' and similar ways of preventing buffer underruns, ide burners were a pain in general (and if you burned enough cds, a scsi burner would easily earn you back the price for a scsi card by being more reliable and creating fewer coasters as a result)
In other words, I somewhat doubt atapicam was very relevant initially, and by the time it really became relevant, it was there.
Oh, and 1998 is really years ago, could say almost a decade ago even.
ide cd-rom drives have been around for much longer, but didn't and still don't require atapicam (that is unless you want to run tosha or similar software against it)
The filesystem metadata policy I see as screwing over mostly for the use on servers etc; and the point is that users *expect* their filesystem to work. They don't go into a system asking "Hmm, when I save a file here, will it be saved or will it be fucked up in some cases? I'd better check the filesystem documentation. Oh, it says nothing about metadata updates. Let's go see the source code. WHAT? UNORDERED, ASYNCHRONOUS METADATA UPDATES? It's not possible to database semantics over this!!!!"
Well, the thing is that if it is used to store the source of what I am currently developing, I quite hope the filesystem does handle my experiments with the kernel and subsequent crashes or dropping into a debugger without corrupting my files. Of course one would want to store that source code elsewhere, but being able to rely somewhat on your filesystem also implies not having to hunt corrupted files as part of trying to figure out why something all of a sudden broke that was working yesterday and such..
I'm sure I have been unlucky there, but crashes and consequent fs corruption resulting from over-enthousiast users on a multiuser machine made the issue pretty easy to see for me, even when not looking at the source code. Yes, I did manage to end up with a corrupted ffs as well a few times, but by far not as often, and pretty much in all cases due to bugs that have been fixed.
Maybe it matters that among other uses, I tried to make use of it as a platform on which to develop and run a mud, and consequentely had multiple people logging in and compiling stuff and such, but hrm, the prospect of a 'real' multiuser system was one of the things that made unix on a pc such an interesting thing for me.
The increased speed it gives can be a large selling point - it's just that people aren't aware of what they're being screwed over with for that speed.
As for losing data: Well, many of us filesystem and VM hackers in FreeBSD actually ran our devel boxes with async metadata back before soft updates. We knew the rough odds, and knew what would happen on a failure. Mail and database servers is what this is critical for - if we got a few fucked up files, we could generally recover from it.
I understand that, but see above. I have used async for filesystems as well for the performance increase it gives, but usually in cases where recreating the data was a minor or non issue (/tmp for example)
That doesn't change that all it would have taken to eliminate this difference was adding an option to the installer so a new user didn't have to hunt documentation for getting the performance increase when desired. I somehow doubt it explains why Linux has been so much more popular then FreeBSD.
As for PnP on 3c509, that's not quite PnP, and it was also imp
I believe running on developer's desktops is most important.
And that is where we differ. I believe that for a system to become practically usable, it needs to get used beyond the developers desktop (unless of course you were building it as a development platform to begin with). Ie, I'm pretty sure there are people using AmigaOS 4 on their development workstation, I'm also pretty sure that those are AmigaOS developers, and that the relevance of that for the rest of the world is virtually zero.
If it can't run either my general purpose workstations, servers, nor on my embedded devices, then an OS has little practical use for me, and I honestly believe that is true for most of the market.
This means that low end hardware must be supported (and Linux was better than BSD there for a while)
And that was what my original post replied to, because unless you talk about the days before FreeBSD 1.1, that is definitely not true, and after 2.2 it is at best debatable.
By 1993, a machine with 24mb ram was not extremely unusual, esp. not as a development workstation or server. Actually supporting DMA for memory over 16mb without serious performance impact made that lots of the typical expansion cards kept working on a FreeBSD system where they'd break or run with a severe performance penalty on Linux.
Earlier you specifically mentioned IDE. I'd be interested in what you are thinking about there. atapicam has been around for quite some time in order to allow usage of ide cd writers and similar devices by attaching them to the scsi cam layer, functionaly similar to scsi over ide as Linux used to have it for the same reasons.
IDE CD rom? No idea, but they seem to have worked for almost as long as I have been able to buy them. It may well be that I am not aware of lack of support of certain chipsets, and I bet there have been quite a few times where Linux happened to support a new EIDE feature or extension, but this never has been an issue that I could detect, this despite having supported lots of BSD as well as Linux installs for more then a decade (currently I support a few dozen FreeBSD desktop installs and a bunch of Linux and FreeBSD based servers)
FreeBSD has always given me far less trouble with isa pnp cards, esp. pretty popular ones like the 3c509 and similar nics.
, and it means that low file system latency is important.
If the price for that is data corruption, then that sounds really like a bad idea on esp. a development workstation used for developing the exact OS that it is running on...
The Linux default for filesystem updates was faster than the BSD default for a long while, because Linux did the choice of running async metadata updates, which violates POSIX and can lose user data but runs faster.
And is quite an option on FreeBSD if you really need that compromise.
I had a couple of fights with Linus over this - basically, I don't see it as OK to fuck over users' data by default because it gives you better benchmarks, and that was actually Linus original primary argument ("We get such lousy benchmarks if we do it right", paraphrased). This is also one of the reasons I'm less than impressed with Linus.
But you still said it was an important argument for its use on a developers workstation?
Back to primary topic: Scheduler quality and VM quality isn't that important in the desktop context,
Hmm.. so that is why people are still pursuing ULE then?
I'm sorry but I believe you are really mistaken about the importance of this for the desktop. You are however right that getting those right is in itself not enough.
Considering that multimedia and voip and video conferencing have become a substantial part of what personal workstations are being used for, for that market having a better scheduler really makes a difference, given all other things are equal.
and while we (BSD) beat Linux on that for a long while (with 2.6, I think Linux has finally gotten
A Linksys WRT54G v2.2, running OpenWRT with the OpenVPN package installed.
No doubt DD-WRT supports this as well.
The point of my post however is that a 'hardware firewall' is just another machine, it is nothing special. You could use a Linksys WRT54G or something compatible, which has some nice features (programmable switch with vlan support and hardware support for AES being the more obvious ones to me, but you could also use a small Soekris box and install for example MiniBSD on it and build the thing from scratch.
Either way you get a tiny, silent, dedicated box without any moving parts that implements your firewall.
Hmm.. back when I used to work there it was usually "I've Been Moved", refering to the yearly seemingly random redistribution of people throughout the company that was policy at the time.
It also helps that IBM cannot possibly block the internet to the 40% of employees who are mobile or at home workers.
In 1990, I was one of the few employees at IBM having internet access from my desk, that is, I had an account on the corporate usenet, smtp, telnet and ftp gateways.
By 1994, there were coporate http and socks 5 proxies, and those were soon accessable to everyone on the company network.
Around the same time, I was involved in one of their moves to get employees more independent from their workplace, and the move towards the 'home office' for a substantial number of employees. (I still have the isdn line from that time, I have to pay for it, but the conditions that apply to it for the telco date back to that time, and are not of the kind you normally get as a private person.. try 24/7 onsite support with a 4 hours guaranteed response time).
Anyway, they had internet access for virtually all employees way before they had a substantial number of people working from home or mobile offices.
While they used to have blocks on certain phone numbers, they seem to be quite trusting of their employees when it comes to things like internet access, and in general, doing your job. While I am no longer working for them (left there some 5 years ago) due to a dispute about work, they did earn enough loyality from me that they didn't need non-disclosure agreements and the like to keep me from going to their competition afterward and releasing what I knew about their internal stuff as a result.
I think it is one of those things that they do get right there.
Show me an employer who places indiscriminate blocks on numbers that you can call during the day, in order to prevent you from making calls that *might* be personal.
One of my previous employers happens to be IBM. They used to have a block on any 0900 (pay) numbers over where I live (the Netherlands). You could call one only by putting in a specific, motivated request.
It wouldn't be very difficult for a net cafe owner to set up an MIM attack and have their own self-signed certificate. Your browser *should* throw a warning, but most users will happily accept the extra risk without thinking twice (or even reading the error message).
And since the owner of that net cafe would have full control over those browsers, it wouldn't be too difficult to install their own 'root certificate' and get rid of the warnings.
You'd have to take a close look at the certificate to spot this.
But solving that problem is a few dollars away in the form of a screen protector.
Which solves only half of the problem of course, people can still easily observe and record your typing.
For the technically uninformed that believe the internet is inherently safe to surf and operate on this article may come as a surprise.
True.
What worries me more is the fact that people regard personal/delicate information as just "something they work with". Reminds me of the day we found social security numbers and copies of military orders in the dumpster at my former Air Force Base. Some people are clueless.
Well, my point was that working with sensitive data in public places is usually a hard to miss sign of such cluelessness for reasons much simpler then that a wireless network might be sniffed.
I'm arguing that the kernel, by itself, is insufficient to operate the computer - to give it commands and have it execute them.
Yes, I understand that, but I am telling you that what you are talking about is called user interface, not operating system. Many operating systems come with one bundeled or even built in, but in no way is it a mandatory part of an operating system.
The Linux kernel, with on top of it a userland consisting of one single, statically linked application, perfectly fits the requirements of an operating system. It allows the application to operate the computer.
Operating system is indeed based on the verb to operate, but not as in having a human 'operate' the machine, but in making the machine able to perform tasks, making the machine operate.
Ok, did my research a bit this time, memory is nice but not always accurate.
The first release of Linux was in mid 1991, while the firt semi-usable release of the 386BSD project (predecessor to Free/NetBSD) was in mid 1992. And I remember using pre-1.0 versions of Linux for at least a year prior to 1.0, and having it feel like a full Unix system.
0.9x till 1.0 lasted a long time, and yeah, 0.95 and later were definitely usable for research and hobby, and definitely felt like a Unix system. That said, I have tried to run a multiuser machine with a bunch of terminals connected to share an internet connection, and while that worked well for users who tried to behave on the system, it was really not yet upto the task for such things (no way of stopping a single user from hogging the machine or even bringing it down at times for example). This was post 0.95, but I do no longer know which version exactly. There is a bit more then 'looks and feels like Unix' to actually acting like a somewhat well developed one (and how could it, it was mostly a 'scratch an itch' derived kernel that was rapidly developing maybe, but only had done so for a short time.
That said, the first Linux version which I consider fully usable for most things for which I'd want a Unix like environment is 2.0. The first 'free / net or open' bsd release that I'd consider fully usable would probably be a NetBSD 0.9 or such, but definitely FreeBSD 1.1. From there till around the time of Linux 2.2, it seems to be a pretty easy call which systems provided a more complete and usable system and a better kernel.
Anyway, I don't think either of us will change our mind on this topic, so it's probably as well to just go on with our lives:-) In peace, Eivind.
I won't. I have used both intensively since their early incarnations. I do think however that it became more clear why we differ in opinion, never bad to have a bit of discussion even if you can't agree.
fraud is a problem, for sure, but mainly for clueless newbs.
Spoken like a true experienced ebayer...
their feedback system has flaws, yes, but it also works. i've bought over $100k of electronics on ebay this year, haven't gotten scammed at all.
Absolutely, it works for you so it must work..
I have never gotten scammed on ebay either, but I only have a few transactions there. My girlfriend however has hundreds of transactions, and only ran into a few issues that usually got resolved. That said, she has decided to move her selling somewhere else. Selling prices are much better, no ever increasing fees, and no organisation that scares away 'clueless newbies' (let me give you a hint, for a seller those are potential new customers), and far less efford.
True. But do you know what statically linking actually does ? It places copies of all the libraries used by the program into the program executable.
I know that..but do you know that there are libraries statically linked into the kernel as well?
Your single program still uses those libraries; they have simply been packed into the same file as the main program.
And they only serve that main program, effectively having become part of it, just like libraries statically linked into the kernel.
In any case, in your example, the userland is the program, and my argument that the kernel by itself is insufficient to make a working OS still stands.
If you want to argue that an OS is usefull withotu any applications then your argument still stands. If you want to hold that MS Office for example is part of the OS for an office worker (since it is what makes a Windows computer 'usable' to them) then your argument stands.. but if you don't think those are true, then I suggest you reconsider your argument and consider that you are talking about different things here, ie:
1. An operating system 2. A user interface 3. An application programming interface 4. An application
1. does implement 3. by definition, but not 2 and 4
The BSD options were not reasonably available at the time Linux gained a foothold, due to licensing problems with AT&T.
This is not true. FreeBSD 2.x was readily available and had been for some time when Linus released the 1.0 kernel. Of course there were kernels before 1.0, and they were even being used and distributed (I still have a Snow CD based on Linux 0.9 kernel around for example), but fact is that the legal issues for BSD got clear before Linux had become anywhere near the usable system it is now.
In addition, I believe there were some issues with hardware support, especially IDE support AFAIR.
Aha? actually, hardware support in at least FreeBSD 2 was a lot better then it was in Linux versions at the time, esp. on systems with more then 16mb memory.
And, not least, the BSDs were fairly complete in themselves, and didn't have any reason to coopt the GNU system - which missed mainly a kernel.
While the resultant settlement left BSD in the clear, at the time the linux kernel was viewed as being safer. The fact that the linux kernel was released under the GPL also meant a lot to GNU-minded individuals
At the time the BSD legal issues were cleared, the Linux kernel was far from where it is now, and BSD still provided a clearly superior system.
Also, after the legal issues were cleared, the resulting clarity could have achieved the exact opposite from what happened initially, and people could have felt more safe using it.
Legal issues does explain part of why people were carefull at first, but it doesn't explain the entire story by far.
Apparently ebay already disapproves of this practice, but obviously doesn't do anything about it.
Not only don't they do anything about it, they effectively encourage it.
Selling a $1 item with $30 postage costs a fraction in ebay fees compared to selling the item for $25 with $6 postage (depending however on where you are selling from however, in my home country Ebay never got a foothold due to a very effective local service (owned by ebay now as well, surprise surprise) and they use fixed fees for auctions)
Not only do I as an eBay sell find that I often have to relist items to sell to a paying buyer, but I also was upset to learn that eBay has been sending out emails trying to convince its users to support the evil that is net neutrality.
1. this is irrelevant to the issue 2. you may want to get a clue about what net neutrality means instead of reading the absurd claims from extremists
It is not the whole system by itself since, by itself, it is insufficient to operate the system. Just try it: install Lilo (or Grub) and Linux kernel to an empty partition, and try to boot. The kernel will halt with kernel panic, since it can't find init, and can't do anything useful by itself.
But I could add a single application, name it init, and as long as it is linked statically to everything else it might need, this would work fine.
userland provides lots of libraries and tools, and with that, an application and user interface. It is not needed for running a specific (and correctly linked) application however.
And, if my experience is any indication, the reason people picked Linux over BSD is the userland (and maybe the license): until a few years ago, the BSD kernel was better than the Linux kernel, but the BSD userland has always sucked compared to the GNU userland.
This might have been true 10 years ago, and maybe even 5 years ago, but it is definitely not true today.
First of all, there is very little in userland programs that is available on Linux but not on BSD. Then, where there are differences between the two, they usually show a better and esp. more workable implementation on BSD. For a simple example, compare the typical ftp client you find on Linux to the ftp client you find on for example Net or FreeBSD..
Then, the BSD userland has a much cleaner, easier to follow directory layout.
And last but not least, in most cases, you can easily compile a Linux userland tool on a BSD and have it work.
Now looking at the kernel, the differences are substantial in things that actually matter to end-users. Sure, BSDs support usb, but it is quite a hastle to make anything complex work, and don't even talk about things like full-duplex audio over usb. Where is a camera usb class? When can I start using the bi-directional connection with my printer to actually get some info from it and make its integrated scanner work? (and yeah, there is a new usb2 project to address those specific issues, but outside any of the 'official' source trees). How about wireless networking? (it is quite there now in FreeBSD 6.x, but that is really really recent)..
Fixable? most of it probably is, but there is one, and one reason only why I have a Linux install next to lots of BSD installs here, and it is things in the Linux kernel that matter for workstation use that I cannot get at all or only at great difficulty on a BSD system.
I think your experience is also fairly acedemic, and realities of actually finishing something to the point where others get interested in it, and then 'selling' it well so people keep being interested in it and actually put in some work as well is what the big deal is here.
Linux' kernel wasn't the only 'good enough' option at the time, and in quite a few cases it was for quite some time not good enough. Rather, for many purposes, for quite some time the BSD variations running on i386 were a much more complete and better option or unless you wanted to put in a lot of efford the only option.
So.. while your reasoning follows logic to explain away the relevance of what Linus achieved, it ignores something called reality I believe.
Slashdot Mirror
FreeBSD's 7 arches probably cover 99% of the computers people would want to use.
.....
Incidentely, of the 7 computers I have at home here, there are 5 where FreeBSD 7 does not run, and 4 on which no version of FreeBSD can boot even in theory.. Those are 2 HP/9000 systems, a Sun Sparcserver 20, an Amiga 3000, and a slightly older x86 machine. 2 of those are in use everyday (the sparcserver is internal dns and the older x86 machine is a slightly 'advanced' x terminal.
While NetBSD has sacrificed features and speed for portability, FreeBSD has managed most of the portability (from a practical standpoint)
No they have not. FreeBSD 7 does indeed run on many fairly modern machines, but does not run well on anything a few years old regardless of what cpu and architecture it uses.
What NetBSD's 'portability' results in, and which you seem to not realize at all, is in much cleaner solutions that will not break on some obscure, seemingly unrelated changes. You may not realize how important that is untill you tried to write an OS yourself maybe..
while adding new features.
Lets see... FreeBSD's usb support comes from NetBSD, so does its bluetooth support. Ah, but FreeBSD did add wireless and acpi support more decently during that same time..
Seems like a pretty even score to me actually, even while I am sure I forgot some things on both sides.
OpenBSD has a good niche, as security is a goal for while people are willing to sacrifice some features and speed. Portability alone is a strange goal however, since the only question that really matters is "does it run on all the computers I have".
"I want to use the same base system on all hardware that I have" is not strange at all. You obviously never have been in a large environment where different kinds of hardware are used because they are good at different kinds of things, but people would still want to run the same basic OS on them. Not everything is a PC, and that is good because PCs really suck at some things.
I'm a Linux user myself, so I don't have much reason for favoritism for a particular flavor of BSD. I try to keep up with OS news in general however.
I don't have anything against NetBSD, though I can't say I'd miss it that much (just like slackware as a Linux distro - historically important, but now largely irrelevant).
And that is why I will now say that you really have very little understanding of what NetBSD, FreeBSD, OpenBSD and the like are doing, about their similarities and differences, and what the relevance of NetBSD is as a system. It is sadly mostly irrelevant from a user point of view, it is definitely not from an OS developer point of view, and as such it has quite some relevance for even those who develop Linux right now.
If I had any current concern, it would be that I really hope DragonflyBSD succeeds, as it is pursuing some really interesting ideas in modern OS design.
If you'd have read the article and understood it somewhat, you'd realize that if anything, Dragonfly suffers from some of the exact same issues that resulted in the current state of NetBSD, and from my point of view, their relevance is similar, relevant to OS developers, not to end-users. That however does not mean that there is no reason for it to exist, and I fail to see any argument in your post suggesting why that is different for NetBSD.
I once by curiosity found this excellent article describing AS/400 pricing systems.
:) ), and I guess IBM still has a very good salesforce to get their stuff out, but how long can they keep this up with pricing systems like this?
AS/400s are mini computers, not mainframes..
And I was shocked. If you buy the low-end version, you already have the full power chip, but just slowed down. For about $ 20k per processor you can unlock that, but that will also require you to buy another license, for several additional k$ per processor.
Usually that also comes with on-site service and the like of a kind that Dell and friends haven't heard of yet.
Also check the price/performance comparison for competing UNIX and Linux systems shown there. What kind of pricing system is that?
I don't know this particular study, but I'd mostly be looking for an AS/400 if I'm looking for something to run a huge database, not hundreds of virtual machines.
I can imagine the hardware is a bit more stable etc, but relative performance of the Power PC is not as good it used to be.
The performance of a machine depends on a lot more then the CPU however. IO bandwidth is a major factor in performance, esp. when the dataset you work on gets substantially bigger then fits in cache or even main memory.
Hence looking at the cpu architecture is not going to tell you much about a system which specializes in IO bandwidth (mainframe class computers)
Stability of hardware, possibility to replace hardware without taking the system down and such are also part of the picture of course, but not unique to mainframe hardware.
Still IBM is probably just asking prices that it thinks it can get away with. Why are they doing this?
Uh, because they are a business and are in it for the money?
And, wouldn't someone else be able to use commodity chips (some workstation-grade core 2 duo or opteron) and build a robust system out of that, for only part of they price they ask.
Seeing the amounts of money going around in that market, someone would have tried.. As a matter of fact, people tried more then once, and in the mid 90s the mainframe market looked dead and it looked like commodity hardware was outdating it. Somehow, this never really happened however.
Thomas J. Watson was a very good salesman (a bit too good to be legal
Untill someone beats them at it or the market just dissapears.. Right now however it looks like they beat everyone who tried to beat them.
Welcome to the real world where businesses exist to 'produce shareholder value', or more generally, are in it for the money.
Uhm, you're mixing timelines again. CAM wasn't until 1998 (or so) and replaced the old SCSI subsystem, ATAPICAM came after that again. And the ULE scheduler is important TODAY, while I'm talking about what was important in 1993-1995.
I remember buying my last scsi cd writer in late 1998 (a 4 speed plexwriter) which was actually slightly cheaper then an equivalent quality ide writer at the time. The same was true for the phillips cd writer I had before that.
It meant having to have a scsi controller, sure, but those didn't cost the world either, and untill the widespread use of 'burnfree' and similar ways of preventing buffer underruns, ide burners were a pain in general (and if you burned enough cds, a scsi burner would easily earn you back the price for a scsi card by being more reliable and creating fewer coasters as a result)
In other words, I somewhat doubt atapicam was very relevant initially, and by the time it really became relevant, it was there.
Oh, and 1998 is really years ago, could say almost a decade ago even.
ide cd-rom drives have been around for much longer, but didn't and still don't require atapicam (that is unless you want to run tosha or similar software against it)
The filesystem metadata policy I see as screwing over mostly for the use on servers etc; and the point is that users *expect* their filesystem to work. They don't go into a system asking "Hmm, when I save a file here, will it be saved or will it be fucked up in some cases? I'd better check the filesystem documentation. Oh, it says nothing about metadata updates. Let's go see the source code. WHAT? UNORDERED, ASYNCHRONOUS METADATA UPDATES? It's not possible to database semantics over this!!!!"
Well, the thing is that if it is used to store the source of what I am currently developing, I quite hope the filesystem does handle my experiments with the kernel and subsequent crashes or dropping into a debugger without corrupting my files. Of course one would want to store that source code elsewhere, but being able to rely somewhat on your filesystem also implies not having to hunt corrupted files as part of trying to figure out why something all of a sudden broke that was working yesterday and such..
I'm sure I have been unlucky there, but crashes and consequent fs corruption resulting from over-enthousiast users on a multiuser machine made the issue pretty easy to see for me, even when not looking at the source code. Yes, I did manage to end up with a corrupted ffs as well a few times, but by far not as often, and pretty much in all cases due to bugs that have been fixed.
Maybe it matters that among other uses, I tried to make use of it as a platform on which to develop and run a mud, and consequentely had multiple people logging in and compiling stuff and such, but hrm, the prospect of a 'real' multiuser system was one of the things that made unix on a pc such an interesting thing for me.
The increased speed it gives can be a large selling point - it's just that people aren't aware of what they're being screwed over with for that speed.
As for losing data: Well, many of us filesystem and VM hackers in FreeBSD actually ran our devel boxes with async metadata back before soft updates. We knew the rough odds, and knew what would happen on a failure. Mail and database servers is what this is critical for - if we got a few fucked up files, we could generally recover from it.
I understand that, but see above. I have used async for filesystems as well for the performance increase it gives, but usually in cases where recreating the data was a minor or non issue (/tmp for example)
That doesn't change that all it would have taken to eliminate this difference was adding an option to the installer so a new user didn't have to hunt documentation for getting the performance increase when desired. I somehow doubt it explains why Linux has been so much more popular then FreeBSD.
As for PnP on 3c509, that's not quite PnP, and it was also imp
I believe running on developer's desktops is most important.
And that is where we differ. I believe that for a system to become practically usable, it needs to get used beyond the developers desktop (unless of course you were building it as a development platform to begin with). Ie, I'm pretty sure there are people using AmigaOS 4 on their development workstation, I'm also pretty sure that those are AmigaOS developers, and that the relevance of that for the rest of the world is virtually zero.
If it can't run either my general purpose workstations, servers, nor on my embedded devices, then an OS has little practical use for me, and I honestly believe that is true for most of the market.
This means that low end hardware must be supported (and Linux was better than BSD there for a while)
And that was what my original post replied to, because unless you talk about the days before FreeBSD 1.1, that is definitely not true, and after 2.2 it is at best debatable.
By 1993, a machine with 24mb ram was not extremely unusual, esp. not as a development workstation or server. Actually supporting DMA for memory over 16mb without serious performance impact made that lots of the typical expansion cards kept working on a FreeBSD system where they'd break or run with a severe performance penalty on Linux.
Earlier you specifically mentioned IDE. I'd be interested in what you are thinking about there. atapicam has been around for quite some time in order to allow usage of ide cd writers and similar devices by attaching them to the scsi cam layer, functionaly similar to scsi over ide as Linux used to have it for the same reasons.
IDE CD rom? No idea, but they seem to have worked for almost as long as I have been able to buy them. It may well be that I am not aware of lack of support of certain chipsets, and I bet there have been quite a few times where Linux happened to support a new EIDE feature or extension, but this never has been an issue that I could detect, this despite having supported lots of BSD as well as Linux installs for more then a decade (currently I support a few dozen FreeBSD desktop installs and a bunch of Linux and FreeBSD based servers)
FreeBSD has always given me far less trouble with isa pnp cards, esp. pretty popular ones like the 3c509 and similar nics.
, and it means that low file system latency is important.
If the price for that is data corruption, then that sounds really like a bad idea on esp. a development workstation used for developing the exact OS that it is running on...
The Linux default for filesystem updates was faster than the BSD default for a long while, because Linux did the choice of running async metadata updates, which violates POSIX and can lose user data but runs faster.
And is quite an option on FreeBSD if you really need that compromise.
I had a couple of fights with Linus over this - basically, I don't see it as OK to fuck over users' data by default because it gives you better benchmarks, and that was actually Linus original primary argument ("We get such lousy benchmarks if we do it right", paraphrased). This is also one of the reasons I'm less than impressed with Linus.
But you still said it was an important argument for its use on a developers workstation?
Back to primary topic: Scheduler quality and VM quality isn't that important in the desktop context,
Hmm.. so that is why people are still pursuing ULE then?
I'm sorry but I believe you are really mistaken about the importance of this for the desktop. You are however right that getting those right is in itself not enough.
Considering that multimedia and voip and video conferencing have become a substantial part of what personal workstations are being used for, for that market having a better scheduler really makes a difference, given all other things are equal.
and while we (BSD) beat Linux on that for a long while (with 2.6, I think Linux has finally gotten
A Linksys WRT54G v2.2, running OpenWRT with the OpenVPN package installed.
No doubt DD-WRT supports this as well.
The point of my post however is that a 'hardware firewall' is just another machine, it is nothing special. You could use a Linksys WRT54G or something compatible, which has some nice features (programmable switch with vlan support and hardware support for AES being the more obvious ones to me, but you could also use a small Soekris box and install for example MiniBSD on it and build the thing from scratch.
Either way you get a tiny, silent, dedicated box without any moving parts that implements your firewall.
Sure you can use a hardware packet firewall for the essential functionality
And much more, depending on what you buy of course
, but what of detailed logs,
My linksys 'hardware firewall' runs a real syslog, and produces a nice duplicate of its log on another machine.
intrusion detection software,
If it isn't too big and you have the source.. my 'box' happens to run a simple snort configuration.
VPN solutions,
It also supports ipsec and as a nice bonus openvpn
SSH proxying, etc.?
Not just that, I can ssh into it and have a nice prompt..
Aside from that, the hardware firewalls can be cracked, and have been in the past.
Same for 'software firewalls'
But they're harder to upgrade and repair when an OSS patch is released a few hours later.
Now here is some news, you can have both.
You are of course quite correct about the faith people put in firewalls, they are a tool, not an all covering security solution.
Hmm.. back when I used to work there it was usually "I've Been Moved", refering to the yearly seemingly random redistribution of people throughout the company that was policy at the time.
I somehow think that was the point of my post indeed :)
It also helps that IBM cannot possibly block the internet to the 40% of employees who are mobile or at home workers.
In 1990, I was one of the few employees at IBM having internet access from my desk, that is, I had an account on the corporate usenet, smtp, telnet and ftp gateways.
By 1994, there were coporate http and socks 5 proxies, and those were soon accessable to everyone on the company network.
Around the same time, I was involved in one of their moves to get employees more independent from their workplace, and the move towards the 'home office' for a substantial number of employees. (I still have the isdn line from that time, I have to pay for it, but the conditions that apply to it for the telco date back to that time, and are not of the kind you normally get as a private person.. try 24/7 onsite support with a 4 hours guaranteed response time).
Anyway, they had internet access for virtually all employees way before they had a substantial number of people working from home or mobile offices.
While they used to have blocks on certain phone numbers, they seem to be quite trusting of their employees when it comes to things like internet access, and in general, doing your job. While I am no longer working for them (left there some 5 years ago) due to a dispute about work, they did earn enough loyality from me that they didn't need non-disclosure agreements and the like to keep me from going to their competition afterward and releasing what I knew about their internal stuff as a result.
I think it is one of those things that they do get right there.
Show me an employer who places indiscriminate blocks on numbers that you can call during the day, in order to prevent you from making calls that *might* be personal.
One of my previous employers happens to be IBM. They used to have a block on any 0900 (pay) numbers over where I live (the Netherlands). You could call one only by putting in a specific, motivated request.
I didn't see a problem with that btw.
It wouldn't be very difficult for a net cafe owner to set up an MIM attack and have their own self-signed certificate. Your browser *should* throw a warning, but most users will happily accept the extra risk without thinking twice (or even reading the error message).
And since the owner of that net cafe would have full control over those browsers, it wouldn't be too difficult to install their own 'root certificate' and get rid of the warnings.
You'd have to take a close look at the certificate to spot this.
But solving that problem is a few dollars away in the form of a screen protector.
Which solves only half of the problem of course, people can still easily observe and record your typing.
For the technically uninformed that believe the internet is inherently safe to surf and operate on this article may come as a surprise.
True.
What worries me more is the fact that people regard personal/delicate information as just "something they work with". Reminds me of the day we found social security numbers and copies of military orders in the dumpster at my former Air Force Base. Some people are clueless.
Well, my point was that working with sensitive data in public places is usually a hard to miss sign of such cluelessness for reasons much simpler then that a wireless network might be sniffed.
Say a Microsoft Word document gets transmitted. The sniffer program will collect that and someone could open it up on their computer
Yeah, but while in a public place, someone looking over your shoulder might be a more realistic worry.
I'm arguing that the kernel, by itself, is insufficient to operate the computer - to give it commands and have it execute them.
Yes, I understand that, but I am telling you that what you are talking about is called user interface, not operating system. Many operating systems come with one bundeled or even built in, but in no way is it a mandatory part of an operating system.
The Linux kernel, with on top of it a userland consisting of one single, statically linked application, perfectly fits the requirements of an operating system. It allows the application to operate the computer.
Operating system is indeed based on the verb to operate, but not as in having a human 'operate' the machine, but in making the machine able to perform tasks, making the machine operate.
Ok, did my research a bit this time, memory is nice but not always accurate.
:-) In peace, Eivind.
The first release of Linux was in mid 1991, while the firt semi-usable release of the 386BSD project (predecessor to Free/NetBSD) was in mid 1992. And I remember using pre-1.0 versions of Linux for at least a year prior to 1.0, and having it feel like a full Unix system.
0.9x till 1.0 lasted a long time, and yeah, 0.95 and later were definitely usable for research and hobby, and definitely felt like a Unix system. That said, I have tried to run a multiuser machine with a bunch of terminals connected to share an internet connection, and while that worked well for users who tried to behave on the system, it was really not yet upto the task for such things (no way of stopping a single user from hogging the machine or even bringing it down at times for example). This was post 0.95, but I do no longer know which version exactly. There is a bit more then 'looks and feels like Unix' to actually acting like a somewhat well developed one (and how could it, it was mostly a 'scratch an itch' derived kernel that was rapidly developing maybe, but only had done so for a short time.
That said, the first Linux version which I consider fully usable for most things for which I'd want a Unix like environment is 2.0. The first 'free / net or open' bsd release that I'd consider fully usable would probably be a NetBSD 0.9 or such, but definitely FreeBSD 1.1. From there till around the time of Linux 2.2, it seems to be a pretty easy call which systems provided a more complete and usable system and a better kernel.
Anyway, I don't think either of us will change our mind on this topic, so it's probably as well to just go on with our lives
I won't. I have used both intensively since their early incarnations. I do think however that it became more clear why we differ in opinion, never bad to have a bit of discussion even if you can't agree.
Peace.
fraud is a problem, for sure, but mainly for clueless newbs.
Spoken like a true experienced ebayer...
their feedback system has flaws, yes, but it also works.
i've bought over $100k of electronics on ebay this year, haven't gotten scammed at all.
Absolutely, it works for you so it must work..
I have never gotten scammed on ebay either, but I only have a few transactions there. My girlfriend however has hundreds of transactions, and only ran into a few issues that usually got resolved. That said, she has decided to move her selling somewhere else. Selling prices are much better, no ever increasing fees, and no organisation that scares away 'clueless newbies' (let me give you a hint, for a seller those are potential new customers), and far less efford.
True. But do you know what statically linking actually does ? It places copies of all the libraries used by the program into the program executable.
I know that..but do you know that there are libraries statically linked into the kernel as well?
Your single program still uses those libraries; they have simply been packed into the same file as the main program.
And they only serve that main program, effectively having become part of it, just like libraries statically linked into the kernel.
In any case, in your example, the userland is the program, and my argument that the kernel by itself is insufficient to make a working OS still stands.
If you want to argue that an OS is usefull withotu any applications then your argument still stands. If you want to hold that MS Office for example is part of the OS for an office worker (since it is what makes a Windows computer 'usable' to them) then your argument stands.. but if you don't think those are true, then I suggest you reconsider your argument and consider that you are talking about different things here, ie:
1. An operating system
2. A user interface
3. An application programming interface
4. An application
1. does implement 3. by definition, but not 2 and 4
The BSD options were not reasonably available at the time Linux gained a foothold, due to licensing problems with AT&T.
This is not true. FreeBSD 2.x was readily available and had been for some time when Linus released the 1.0 kernel. Of course there were kernels before 1.0, and they were even being used and distributed (I still have a Snow CD based on Linux 0.9 kernel around for example), but fact is that the legal issues for BSD got clear before Linux had become anywhere near the usable system it is now.
In addition, I believe there were some issues with hardware support, especially IDE support AFAIR.
Aha? actually, hardware support in at least FreeBSD 2 was a lot better then it was in Linux versions at the time, esp. on systems with more then 16mb memory.
And, not least, the BSDs were fairly complete in themselves, and didn't have any reason to coopt the GNU system - which missed mainly a kernel.
That of course is quite true.
While the resultant settlement left BSD in the clear, at the time the linux kernel was viewed as being safer. The fact that the linux kernel was released under the GPL also meant a lot to GNU-minded individuals
At the time the BSD legal issues were cleared, the Linux kernel was far from where it is now, and BSD still provided a clearly superior system.
Also, after the legal issues were cleared, the resulting clarity could have achieved the exact opposite from what happened initially, and people could have felt more safe using it.
Legal issues does explain part of why people were carefull at first, but it doesn't explain the entire story by far.
Apparently ebay already disapproves of this practice, but obviously doesn't do anything about it.
Not only don't they do anything about it, they effectively encourage it.
Selling a $1 item with $30 postage costs a fraction in ebay fees compared to selling the item for $25 with $6 postage (depending however on where you are selling from however, in my home country Ebay never got a foothold due to a very effective local service (owned by ebay now as well, surprise surprise) and they use fixed fees for auctions)
Not only do I as an eBay sell find that I often have to relist items to sell to a paying buyer, but I also was upset to learn that eBay has been sending out emails trying to convince its users to support the evil that is net neutrality.
1. this is irrelevant to the issue
2. you may want to get a clue about what net neutrality means instead of reading the absurd claims from extremists
It is not the whole system by itself since, by itself, it is insufficient to operate the system. Just try it: install Lilo (or Grub) and Linux kernel to an empty partition, and try to boot. The kernel will halt with kernel panic, since it can't find init, and can't do anything useful by itself.
But I could add a single application, name it init, and as long as it is linked statically to everything else it might need, this would work fine.
userland provides lots of libraries and tools, and with that, an application and user interface. It is not needed for running a specific (and correctly linked) application however.
And, if my experience is any indication, the reason people picked Linux over BSD is the userland (and maybe the license): until a few years ago, the BSD kernel was better than the Linux kernel, but the BSD userland has always sucked compared to the GNU userland.
This might have been true 10 years ago, and maybe even 5 years ago, but it is definitely not true today.
First of all, there is very little in userland programs that is available on Linux but not on BSD. Then, where there are differences between the two, they usually show a better and esp. more workable implementation on BSD. For a simple example, compare the typical ftp client you find on Linux to the ftp client you find on for example Net or FreeBSD..
Then, the BSD userland has a much cleaner, easier to follow directory layout.
And last but not least, in most cases, you can easily compile a Linux userland tool on a BSD and have it work.
Now looking at the kernel, the differences are substantial in things that actually matter to end-users.
Sure, BSDs support usb, but it is quite a hastle to make anything complex work, and don't even talk about things like full-duplex audio over usb. Where is a camera usb class? When can I start using the bi-directional connection with my printer to actually get some info from it and make its integrated scanner work? (and yeah, there is a new usb2 project to address those specific issues, but outside any of the 'official' source trees). How about wireless networking? (it is quite there now in FreeBSD 6.x, but that is really really recent)..
Fixable? most of it probably is, but there is one, and one reason only why I have a Linux install next to lots of BSD installs here, and it is things in the Linux kernel that matter for workstation use that I cannot get at all or only at great difficulty on a BSD system.
I think your experience is also fairly acedemic, and realities of actually finishing something to the point where others get interested in it, and then 'selling' it well so people keep being interested in it and actually put in some work as well is what the big deal is here.
Linux' kernel wasn't the only 'good enough' option at the time, and in quite a few cases it was for quite some time not good enough. Rather, for many purposes, for quite some time the BSD variations running on i386 were a much more complete and better option or unless you wanted to put in a lot of efford the only option.
So.. while your reasoning follows logic to explain away the relevance of what Linus achieved, it ignores something called reality I believe.
One small addition to your interesting post.
There exist games without music and still images, in fact, many of the early adventure games were entirely text based.