Scott Long announced the availability of FreeBSD 4.10-BETA for i386. 4.10-BETA for alpha will be following shortly as they work out some problems. 4.10 is the next step in the 4-STABLE branch, and as such contains primarily bug fixes and incremental functionality improvements. One significant new feature is the merging of the USB stack and drivers from 5.x.
FreeBSD 4.10-Beta Available for i386
About The Author
Eugenia Loli
Ex-programmer, ex-editor in chief at OSNews.com, now a visual artist/filmmaker.
Follow me on Twitter @EugeniaLoli
57 Comments
I call B.S.
The BSDs do in fact use synchronous file systems while the various Linux distributions DO NOT. That alone makes the BSD’s a better choice if stability is your main concern, amongst other things. Where in the hell do you get your information from, or do you just make this sh*t up? And before you lable me as a BSD freak, I have been using RedHat/Fedora for 2 years now exclusively and have no intention on switching.
Hi
You are misinformed. Setting up syncronous filesystems in linux is just a option in /etc/fstab. If you are using linux for 2 years you should be aware of this
Setting up syncronous filesystems in linux is just a option in /etc/fstab.
Same with the BSDs IIRC, but I believe the point is that by default, most (as in every one I’ve ever tried) Linux distributions ship with async mounts for the speed increase, despite the fact that a crash can really bork your filesystem (less of an issue with journaling, but still an issue), whereas the BSDs have traditionally shipped with sync mounts for stability.
Same with the BSDs IIRC, but I believe the point is that by default, most (as in every one I’ve ever tried) Linux distributions ship with async mounts for the speed increase, despite the fact that a crash can really bork your filesystem (less of an issue with journaling, but still an issue), whereas the BSDs have traditionally shipped with sync mounts for stability.
Err.. it isn’t an issue with journalling. This is what journalling was invented for. FreeBSD nowadays uses async mounts by default too when they have softupdates turned on, IIRC.
Time to put this one to rest, I think. In fact they both use asynchronous filesystems with integrety preserving features nowadays.
That is completely untrue. FFS is used and designed as a sychronous filesystem. Even with the metajournaling of softupdates this is still true. Of course, FFS can be mounted async functioning fine, but it’s not how it was designed to be run. The reverse can be said about the Linux filesystems.
Even if this were true, what is the point? (But it isn’t true. Linux 2.4, 2.5, 2.6 is roughly equivalent to FreeBSD 4-STABLE, 5-CURRENT, and 5-STABLE).
The point is, development of kernel + userland is directly coordinated in each of the BSD’s. Although there is some cooperation between Linux & GNU, they are different groups and obviously not quite as in sync as a single body. This can be the cause of some issues, as it has several times in the past.
The fact that such a large machine runs Linux in production is testament to the stability of its SMP implementation.
Just because it is capable of running such a configuration does not mean it does it with stability & reliability. Also, considering that you are using such a large number of CPU’s, it’s quite clear that the emphasis of this system is on performance rather than stability. Or if you want to argue the number of these CPU’s providing added stability through redundancy, you should take note that Linux doesn’t even hold a candle to Solaris in this respect. Either way, scalability & stability are not the same thing.
No, BSD, being an “operating environment” tries to be everything to everyone. Linux, being a kernel, tries to be a good kernel to everyone.
BSD’s design goals, user base, and strengths are very clear; namely, firstmost looking at stability & reliability with all else coming afterwards. BSD is primarily meant for uniprocessor internet servers, and its centralized development allows it to maintain this focus. The Linux community on the other hand, does not have any single focus. Some want it to be a desktop replacement for Windows, others want it to replace proprietary unices on high end systems. Aside from conflicting focuses, development efforts are also fragmented. This leads to rapid development, but not particularly consistent nor mature code.
Err.. it isn’t an issue with journalling. This is what journalling was invented for. FreeBSD nowadays uses async mounts by default too when they have softupdates turned on, IIRC.
I think you misinterpreted my staement. I meant that journalling (which as you said was designed to help prevent badness from occuring on crashing etc.) is not a complete failsafe. Damage still happens despite journalling, and there is little (that I can think of) that will further mitigate the damage save using a synchronous mount.
I’ll admit that I didn’t actually check that when I tried FreeBSD 5.2.1, but everything was mounted sync in 5.1, right along with softupdates out of the box.
That is completely untrue. FFS is used and designed as a sychronous filesystem. Even with the metajournaling of softupdates this is still true. Of course, FFS can be mounted async functioning fine, but it’s not how it was designed to be run.
Err no it isn’t completely untrue. Softupdates are designed explicitly to allow filesystems to be mounted async without metadata consistiency problems.
The reverse can be said about the Linux filesystems.
Default Linux filesystems for years have all been journalled.
Just because it is capable of running such a configuration does not mean it does it with stability & reliability.
It does mean that Linux’s SMP implementation is stable and reliable, which is what I said.
Also, considering that you are using such a large number of CPU’s, it’s quite clear that the emphasis of this system is on performance rather than stability.
You think so? You are “quite clear” that NASA doesn’t care their system crashes after it is halfway through a month-long simulation? OK I can’t argue against that logic
Or if you want to argue the number of these CPU’s providing added stability through redundancy, you should take note that Linux doesn’t even hold a candle to Solaris in this respect. Either way, scalability & stability are not the same thing.
Err a) why did you bring Solaris into this? b) Solaris has so far proven itself to 144 CPUs, which is a long way off Linux, and c) Of course scalability and stability are not the same thing. But this 512CPU system proves Linux’s SMP implementation is *both* scalable and stable.
I think you misinterpreted my staement. I meant that journalling (which as you said was designed to help prevent badness from occuring on crashing etc.) is not a complete failsafe. Damage still happens despite journalling, and there is little (that I can think of) that will further mitigate the damage save using a synchronous mount.
Oh. Well yeah metadata journalling can still lead to data loss, but ext3 (and I think a couple of other fs’s in Linux) can do full data journalling, which is as safe as a synchronous mount.
I’ll admit that I didn’t actually check that when I tried FreeBSD 5.2.1, but everything was mounted sync in 5.1, right along with softupdates out of the box.
From the handbook (of FFS):
It offers a better way to insure filesystem data integrity, mainly with the “softupdates” option. This option decreases synchronous I/O and increases asynchronous I/O
Although I think they still call it a synch mount, it does actually use asynch IO when using softupdates. Same as a journalling filesystem does synch writes to the journal, for example.
I call B.S.
The BSDs do in fact use synchronous file systems while the various Linux distributions DO NOT.
OK, I didn’t realise they call softupdates mounts sync while still doing async writes to them.
That alone makes the BSD’s a better choice if stability is your main concern, amongst other things.
Well metadata journalling is just as safe as softupdates, while data journalling is safer (as good as sync mounts).
Where in the hell do you get your information from, or do you just make this sh*t up? And before you lable me as a BSD freak, I have been using RedHat/Fedora for 2 years now exclusively and have no intention on switching.
Thats nice for you. That has no relevance to what I’m talking about and I really don’t care either.
Err no it isn’t completely untrue. Softupdates are designed explicitly to allow filesystems to be mounted async without metadata consistiency problems.
I think you need to do some more research on softupdates. Softupdates is essentially a compromise for getting async like performance while maintaining the reliablity of sync. Although both data and metadata writes can be delayed, they are kept ordered thus still guaranteeing a consistent filesystem. Thus, although the writes themselves are async, changes in the filesystem are still sync. The end result is that softupdates provides much stronger reliability than journaling filesystems with comparable performance. FFS can also be mounted as fully async, which is completely different than what softupdates does.
The reverse can be said about the Linux filesystems.
When I said that, I meant that when Linux filesystems are mounted synchronous, they are not functioning as is intended in their design. Thus, they will hardly run optimally as synchronous filesystems.
You think so? You are “quite clear” that NASA doesn’t care their system crashes after it is halfway through a month-long simulation? OK I can’t argue against that logic I think it’s quite clear that the number of CPU’s there is for performance rather for redundancy. Just because performance is the emphasis doesn’t mean that they don’t care about stability; it just means tradeoffs had to be made which lean towards performance.
I still maintain that just because Linux runs on that many cpu’s doesn’t make Linux any more stable; it just means that it has successfully tackled the issues necessary to run on such a system. While this process may have helped to tackle some stability issues along the way, it is far from a final guarantee that the code is completely error free. It is perfectly possible to have an operating system that runs on 512 cpus and that still crashes.
Err a) why did you bring Solaris into this? b) Solaris has so far proven itself to 144 CPUs, which is a long way off Linux, and c) Of course scalability and stability are not the same thing. But this 512CPU system proves Linux’s SMP implementation is *both* scalable and stable.
Solaris has many more features in it to deal with contingencies, end of story. Many of these capabilities are specific to their hardware, but they are in place nonetheless. Linux has barely begun to implement such features, and can not compare to Solaris in this category.
And to some of the other posts out there, no, journaling is not as safe as a synchronous filesystem. It alleviates some of the risk of an asynchronous filesystem but nowhere near all. This additional risk is usually dismissed by the Linux community as a worthwhile tradeoff for performance, but that does not mean it does not exist.
I think you need to do some more research on softupdates. Softupdates is essentially a compromise for getting async like performance while maintaining the reliablity of sync. Although both data and metadata writes can be delayed, they are kept ordered thus still guaranteeing a consistent filesystem. Thus, although the writes themselves are async, changes in the filesystem are still sync.
This is exactly what a metadata journalled filesystem that is mounted “async” does.
The end result is that softupdates provides much stronger reliability than journaling filesystems with comparable performance.
No, not much stronger reliability. Exactly the same. Full data journalling actually provides much stronger guarantees than SU and metadata journalling. Ie. the same as a sync mount.
FFS can also be mounted as fully async, which is completely different than what softupdates does.
Yeah, this is like mounting async without using journalling.
I think it’s quite clear that the number of CPU’s there is for performance rather for redundancy. Just because performance is the emphasis doesn’t mean that they don’t care about stability; it just means tradeoffs had to be made which lean towards performance.
We’re talking about software here, not the hardware. If you think they made software stability tradeoffs then you absolutely wrong.
I still maintain that just because Linux runs on that many cpu’s doesn’t make Linux any more stable;
Of course not. The programmers make it stable. Running Linux there doesn’t change it at all, only prove something about its stability.
[i]it just means that it has successfully tackled the issues necessary to run on such a system. While this process may have helped to tackle some stability issues along the way, it is far from a final guarantee that the code is completely error free. It is perfectly possible to have an operating system that runs on 512 cpus and that still crashes.
Yes it is inevitable that Linux has bugs in it, but once a year races on 16-cpu systems become once an hour races on systems like this so it provides very important test coverage.
And to some of the other posts out there, no, journaling is not as safe as a synchronous filesystem. It alleviates some of the risk of an asynchronous filesystem but nowhere near all. This additional risk is usually dismissed by the Linux community as a worthwhile tradeoff for performance, but that does not mean it does not exist.
Now I suggest you perform some more research. Full data journalling *is* as safe as a synchronous filesystem because writes go synchronously to the journal. And metadata journalling (default for most distros I think) is as safe as softupdates (which is default for FreeBSD nowadays, I think).
When are people suppose to start using 5.x???
For desktop use, 5.x is perfectly usable as it is. 4.x remains the STABLE branch simply because it is already thoroughly tested, and still performs better in certain areas. If you’re running a mission critical server, stick with 4.x. However, for most people’s desktop uses, 5.x is perfectly ready. It will be another year or so however before it becomes convincingly superior to 4.x in performance and tested enough for mission critical use, by BSD standards. By Linux standards, 5.x would have been deemed production ready back in 5.0.
Shortly ^_^
IIRC it is suppose to be happening in a few months.
He says there’s driver support from 5.x – by chance is it ath for my wireless card?
Why is the FreeBSD release engineering going back to the 4.1 codebase? I was under the impression they were already up to 4.9 or so. Were there some problems with the new cooperating multi-tasking scheduler they recently implemented (KSE if I remember correctly)? I hope this isn’t a step backwards for the FreeBSD project, because it seems so far behind comparable “Open Source” projects already.
This is not a step backwards. 5.3 is supposed to be stable. Looks like the 4.x branch is just extended. There actually might be 2 stable released 4.10 and 5.3.
There is a TON of work getting done on 5.3. Just a guess, 5.3 may not be part of the stable branch, it might be postponed until 5.4.
Either case, 5.3 is moving along nicely.
By Linux standards, 5.x would have been deemed production ready back in 5.0.
No, 5.0 had lots of problems and there was (and still is) a big todo list before it goes stable.
The Linux 2.6 development branch was getting huge numbers of stress testers and beta testers before 2.6.0. And 2.6 is very stable, so I don’t know why you would say that. Maybe you are Linus?
There is a TON of work getting done on 5.3. Just a guess, 5.3 may not be part of the stable branch, it might be postponed until 5.4.
You are probably right. There was an interesting thread on the mailing list recently:
http://marc.theaimsgroup.com/?l=freebsd-current&m=108159540221965&w…
PHK (a FreeBSD developer) said 5 won’t be stable for summer. So expect it at the end of 2004 or start of 2005.
Heh … you should read it as “four point ten”, not “four point one zero”.
Just in general, the Linux community is much less rigorous in their testing. There’s just much more emphasis on the newest technology than what’s proven and stable like in the BSD community. Keep in mind 5.x existed a LONG time before 5.0, and was used “widely” (strictly relatively speaking) before 5.0 much longer than 2.6 was in beta. By the time 5.x does actually become the STABLE branch, the level of testing will not even be comparable.
Anyway, my point was mostly that for most purposes, 5.x is ready for use. It’s only still listed as a development branch largely because of BSD’s more conservative approach.
By the time 5.x does actually become the STABLE branch, the level of testing will not even be comparable.
If it’s not considered ‘stable’ by the BSD community, then why has it gone past 5.0? Just logic, but if I didn’t consider a OS to be stable I sure as s*?! wouldn’t brand it as a major release by calling it 5.0.
4.10 beat would be more stable than 5.0 realse?
From Linux kernel 2.4.x to 2.6.x there were considerable performance tunings to increase desktop responsiveness. Do you know if there is anything similar to be expected in FreeBSD 5.x once it becomes the stable branch?
I’m not trying to troll – I know that network server performance is priority for FreeBSD developers. It just happens to be that FreeBSD is, thanks to its integral base system and extensive ports collection, one of the best workstation-desktop solutions available for users who prefer Unix-like operating systems.
>4.10 beat would be more stable than 5.0 realse?
Yes. 4.10-RELEASE is a snapshot from the 4-STABLE branch,
whereas the 5-STABLE branch has not yet been created
(5.0-RELEASE was a ‘developer preview’ based on 5-CURRENT).
The 5-STABLE branch is expected to be created later this
year (‘after summer’ was the last estimate I heard).
–Jon
By Linux standards, 5.x would have been deemed production ready back in 5.0.
It’s sad how true this statement really is…
Not that it really matters but I thought that the next release from 4.x was going to be 4.9.1. What happened to that?
I think FreeBSD 5.x has preempt, at least that thing should help. There’re drivers improvements, etc. And there’s the new scheduler which should work better too.
This, along with the fact that you can use gnome 2.6/kde 3.2/openoffice/mozilla/firefox , makes freebsd a pretty decent desktop platform IMHO.
There was a war against Linux?
This, along with the fact that you can use gnome 2.6/kde 3.2/openoffice/mozilla/firefox , makes freebsd a pretty decent desktop platform IMHO.
I don’t think freebsd is a good desktop solution at all first off all is it’s speed, i try’d archlinux and FreeBSD 5.2.1 on the same machine and Linux is much faster than FreeBSD 5.2.1
I like FreeBSD very much and use it for my servers but on the desktop i use Linux
Even with all kinds off optimalisation settings and recompile off the whole system, linux reacts much faster.
With KDE 3.2 you got the bouncing icon when starting a console it take 1 to 2 bounces in Linux and FreeBSD on the same machine it bounces a lot more 3 to 4 time at least.
FreeBSD 5.2.1 is my one and only OS. I’m using it as desktop, server, everything. I’ve been using exclusively the 5.x branch since 5.0 and I never had any “stability” issues.
The fact is, STABLE should read “I’d bet my life on it” and CURRENT should be “You’ll probably be fine but, hey, shit happens”.
I agree with your analysis of FreeBSD. I have used it for several years now. I really liked the way it was structured and incredibly stable. However, I am a desktop user. Little things caused me a lot of grief. The USB stack not working correctly was a major one. Although it seems to be getting fixed now. Then the amount of hardware support. SATA for instance on the new 5.xx series. Although people are working on that one also.
I have recently switched to Slackware as it is similar to BSD in file structure. It provides the items I was missing in BSD. However, I will miss the beastie. I really miss it now since I am having to learn and re-learn all I have forgotten years ago before I went to BSD.
>Why is the FreeBSD release engineering going back to the 4.1 codebase?
4.10 four dot ten ,as 10 naturally comes after 9(in base 10 math atleast).
4.10 is based the 4.9 codebase.
In my expereance Free BSD was noticably faster than any linux i used when i tried it…I am not bashing linux i use Gentoo at home evryday and love it…but thats what i noticed.
Well, I have to disagree with you saying that linux is still faster. For me FreeBSD on Desktop is noticably faster than linux (KDE3.2.1 on FreeBSD 5.2.1 vs Debian/2.6.5 kernel). Especially things go bad in linux on heavy CPU loads (compiling smth), in FreeBSD I was able to surf web, listen mp3 and no problem, in linux … well … things are far from beeing smooth But still, I think about migrating to linux, because I’m tired to work around all the small things that don’t work in FreeBSD (mounting disks using KDE icons, K3B which I never got working, java VM – to complicated and time consuming). What I can say about FreeBSD is that when it is configured – it is rock stable, pitty is that configuration and fine tuning takes way too much time.
As someone who has used both Slackware and FreeBSD extensively over the past few years, I am just soooo tired of hearing the usual arguments about speed, features, etc… The fact is, they are very close, and generally, it is a hard decision to choose one over the other. I recently had to decide between Slack and FreeBSD on a workstation-oriented project. In the end, I chose FreeBSD, but only by a narrow margin. My reasons for choosing FreeBSD had nothing to do with speed, but with maintainability. Even then, it was a close decision, because Slackware is now getting a fairly decent package system, with swaret.
In general, I found untuned Slackware to be faster than untuned FreeBSD, especially comparing desktop performance with FreeBSD 4.9. But, once I tuned both systems, the performance was quite close (some programs executed faster on one or the other). I think this speaks to the fact that FreeBSD defaults to more conservative settings.
Now, I am well aware that Linux has fixed the “Giant lock” before FreeBSD, and many of the benchmarks show Linux winning, but real life isn’t about benchmarks. In *actual practice*, I have found very little difference. Generally configuration choices and hardware setup make far more difference than whether it is FreeBSD or Linux.
However, the one advantage I notice in actual practice is maintainability and consistency. In FreeBSD, more so than any OS I have ever used, everything seems to be in a logical place, with minimal fuss. Since my time is my most precious commodity, that counts for a lot.
The post you answered wasn’t about FreeBSD vs. Linux comparison, it was answer to my question about how FreeBSD 5.x compares to FreeBSD 4.x from the desktop user’s POV.
But if we start comparing apples with oranges, FreeBSD 4.10 can be compared with some Linux distro that has the latest 2.4.x kernel. And once FreeBSD 5.x becomes officially stable, we’ll see how it compares with some specific Linux distro that uses 2.6.x kernel. Another interesting point of comparison will be DragonFly – once it is officially released.
I’m sorry for going of topic: Does anybody know when GNOME 2.6 will be part af the official ports system? And will I be able to use a simple portupgrade -PP gnome2?
FreeBSD has all ways run nicely for desktop. You have the choice of two schedulers under 5x. X even runs smooth with gtk+ and a pixmap theme under heavy load, minus very slight increase when switching between screens, but nothing truely noticable.
Speed prob, well then that means you did not config it correctly.
It has been in the ports since april 4 (see http://www.freshports.org/x11/gnome2). The binary packages are not on the FreeBSD FTP servers yet, so if you want to use packages to upgrade, look here: http://www.marcuscom.com/tinderbox/. Also, don’t forget to read the upgrade FAQ (http://www.freebsd.org/gnome/docs/faq26.html).
I’ve tried FreeBSD 5.1 on the desktop and I think it’s okay but was let down by lack of native java support and flash plugins for my browsers. They seem to rely on Linux ports for some major packages and I understand it may not be their fault but things like this need sorting out if it is to become as viable as Linux.
The java thing could be inconvenient, but id hardly consider it catastrophic.
As far as flash goes, not having that is like being rid of a pest! :oD
On a serious note though, yes… developers should start supporting the bsd’s with their products (net and open too, not just free)… and stop relying on the linux compatibility layer.
Check the ports, native java exists along with a wrapper for flash. No thinks like like this do not need sorted out to become a viable alternative. It has been and is a viable alternative.
If you want some one to whine at, whine to sun and macromedia or whoever, becuase it is thier fualt for lack of support for the FBSD community.
To those who call 2.6 stable: lies. No distributions means its not in widespread use. Cant be stable then. Second, I have personally stress tested 2.6 and while it is supposedly more scaleable, I have no interest in using it with my products because it has regressions in performance in places that matter to me. I dont care about mouses, mozilla and media players. About FreeBSD 4.10. Nothing, save Debian, can approach thee solidity, coherency and documentation of FreeBSD 4.10. League of its own. I suppose RHEL 2.1 is a possible substitute. FreeBSD 5.x (5.2.1 last I checked) is far further along than 2.6 in “battle testing.” Lots of people have synthetic stuff and anectdotes, but because FreeBSD is properly released (in that it has a coherent userland, compiler, c-library and kernel all released as a whole) I would say 5.x is far more viable than say Gentoo 2004.0.
If you rely on Free/Open things for business, you tend to be a bit more unforgiving than believing random synthetic benchmarks and do what everyone has to do, stress test it yourself and benchmark it yourself so you can gague the performance yourself. I’m not lying to myself when I say, 2.6 is not ready, it has performnace regressions despite what you’ve been hearing. FreeBSD 4.10 is the pinnacle of open software design. If you are looking for solid and free and Linux and dont want to pay, try Centos 3. the more people who use it the better because its probably Linux’s last Free hope of being useful in production. Keep in mind, Linux’s appeal was that it was free, and therefore useful for embedded or specialized products, and most distributions are not suited for that either because they cost too much or are desktop oriented.
I’m not just whining at the FreeBSD developers, I understand that they approached SUN to sort the java thing out, but SUN were not very supportful. Like I said I understand that it’s not really the fault of the freebsd developers but those who do not fully support Unix/Linux/BSD.
Once again, this is not meant as bashing, but:
I have noticed that some applictaions performs better under linux bineary mode under FreeBSD then under native linux.
It is nonsense to say FreeBSD is more thoroughly tested than Linux is. Look @ Debian Stable for example. It requires months or even years of testing in order to be included in Stable.
On the desktop FreeBSD performs very well for me, but I find Arch Linux simpler, easier and it has better hardware and multimedia support. I can’t tell the difference in performance, as I do nothing heavy. As a server however, FreeBSD does a great job for me.
I have used a few different FreeBSD releases from 4.7 to 5.2, and to those saying “speed is relatively the same”, its simply not. The command line even seems heavy for every day use. The desktop stuck me at 640×480, which is unacceptable, and dispite trying everything I could think of, this was best case scenario in each case…
Also, I would like to echo what others have said. How can FreeBSD users possibly believe that their favorate operating system is STILL more tested then Linux? Look at all the large companies that are now investing in the kernel, and common sense says these people would never allow sub-par software into production. Last I checked their were over 20 million Linux users, of which only a small amount would have used 2.5, but I will garentee even if only 5% of the total Linux userbase actually tested 2.5, its still vastly more then are testing FreeBSD.
No amount of internal testing can compare with real world usage for testing stability. Simple fact is FreeBSD isn’t used by many people, so there is no way its getting the real world scenario’s thrown at Linux.
I am glad something like FreeBSD exists, but today, Linux is hardly even classifiable as a hobby OS. FreeBSD fulfills that hobbiest need today still, and thats great. Linux is very commercial in today’s world, and no hobbiest OS can compete with anything commercial…
(for those saying “linux is hobbiest” and “linux will never be commercial”, bull, nothing with the backing of companies like IBM, HP, Dell et cetera can be mistaken as hobbiest, no matter what its history is…)
It is nonsense to say FreeBSD is more thoroughly tested than Linux is.
Until Linux as distributed by Linus and his top guys comes complete with a kernel debugger, I’m not going to trust that it’s been more thoroughly tested than FreeBSD, or any other BSD, Windows, or Mac OS X.
A few million pimply faced teenaged boys with no lives and no girlfriends with penguin stickers on their cobbled together boxes can’t be considdered thoroughly tested, by any stretch of the imagination.
Furthermore, the fact that IBM and a few other companies have bought into the Linux craze says nothing of its quality. Better architectures than Linux or BSD could be more easilly maintained without the need of corporate resources.
First allow me to say BS, Linux is just hobbiest with some commercial support. Until the linux devers get thier act together and stop whining, look at all the BS over XF86 4.4, and start working on stuff standardized and put into place a decent packaging system and the like, I don’t see how it can be classified as any thing other than hobbiest.
Actually internal testing can compare rather nicely to real world test. Also not all testing is done internally. If you don’t believe that check and see for your self.
BTW you should not blame FreeBSD for your inability to configure X.
Linux is an operating system while FreeBSD is an operating environment. It’s rather childish having a shouting match over which is best.
You can take the Linux kernel as a starting point and ship whatever software you like with it as a distribution. FreeBSD however, is a whole distribution of (a derivative of) UNIX, not just a kernel.
In terms of stability, I would say that the FreeBSD-STABLE kernel is more stable than the Linux 2.6 kernel (for mission critical uptimes etc) but that’s only really because it has time to prove itself between releases. I don’t have time to see if my desktop machine falls over, because linux kernel releases are often less than a month apart and I pretty much always upgrade it in the hope of improving things each time!
I use Slackware 9.1 as my distribution and Linux 2.6.5 as the kernel. I can’t really prove it is “stable” until I’ve heavily used/tested it over an extensive period of time. With the rapid release of kernels, this isn’t really possible to test.
Linux is not an os while freebsd is an os. GNU/Linux is an os. every part of bsd is made and tested by the FreeBSD Team. I feel bsd to be more logical, everything has its place ex:/usr/local contains all apps, /usr/share/doc contains all documentation, /usr/src/ contains source code /usr/share/doc contains documentation etc. whereas in linux everything gets messed up quite quickly. bsd has the best pkg management system and if something goes wrong when i install soem pkgs i can restore my system by deleting the pkgdb and /usr/local directory. no need to reinstall like linux. moreover i feel bsd is more stable because since its an integral system all possible problems are rectified before release. about freebsd being slow its because of the conservative configuration of bsd whereby the risk of data loss is minimized with a little perfomance trade off, meaning you can pull the plug without having to worry about data corruption. bsd 5 has backround file checking so you won’t be pulled in to a console asking for root password like in linux. benchmarks saying linux is better may be true but when it comes to realtime benchmarking , i find X to be far more responsive in bsd than on linux(i have not tried 2.6). But ofcourse linux is more asy to use than bsd. it takes a long time to setup a bsd system but in return you get a fully optimised system which you can easily upgrade with a few commands.
Sure Linux 2.6 has performance regressions. Things are constantly being evaluated and tradeoffs made. FreeBSD 5 has a huge performance regression over FreeBSD 4 in single threaded uniprocessor applications, for example (yes, even with all debugging stuff off).
So what regressions does 2.6 have in places that matter to you?
As I already mentioned before, Linux & BSD have different design focuses. Linux is more focused on new features & technology; BSD is first and foremost focused on being rock solid reliable. The fact that Linux uses asynchronous filesystems while BSD uses synchronous filesystems is just one example.
Secondly, Linux is only a kernel while the BSD’s are a complete operating systems. While Linux has to deal with multiple permutations, each BSD only has to deal with one at any given time.
When you mention the 512 CPU scalability, that has nothing to do with stability & reliability. In fact, meeting these scalability goals means bringing in new code. New code needs new testing.
And if you really want to look at some benchmarks at stability, have a look at this: http://uptime.netcraft.com/up/today/top.avg.html
Both Linux & BSD have their advantages. Thorough testing goes to BSD, simply due to its more conservative design approach. BSD has a clear, concise focus whereas Linux tries to be everything to everyone. Whether this is positive or negative has to do with what you’re using it for. If you have a 512 CPU system, you’ll clearly want to go with Linux. For another system, a different operating system may or may not be more appropriate. Quite simply, use the best tool for the job instead of trying to argue the universal applicability of one tool for all jobs.
My original post had nothing to do with bashing by the way; I was simply using an analogy to explain the code maturity of the CURRENT branch.
As I already mentioned before, Linux & BSD have different design focuses. Linux is more focused on new features & technology; BSD is first and foremost focused on being rock solid reliable. The fact that Linux uses asynchronous filesystems while BSD uses synchronous filesystems is just one example.
Time to put this one to rest, I think. In fact they both use asynchronous filesystems with integrety preserving features nowadays.
Secondly, Linux is only a kernel while the BSD’s are a complete operating systems. While Linux has to deal with multiple permutations, each BSD only has to deal with one at any given time.
Even if this were true, what is the point? (But it isn’t true. Linux 2.4, 2.5, 2.6 is roughly equivalent to FreeBSD 4-STABLE, 5-CURRENT, and 5-STABLE).
When you mention the 512 CPU scalability, that has nothing to do with stability & reliability. In fact, meeting these scalability goals means bringing in new code. New code needs new testing.
Very little new code will be needed with 2.6 actually. But in fact the sheer parallelism, huge throughput, memory, IO, etc. is actually very effective at finding obscure races and bugs that would otherwise be much rarer.
The fact that such a large machine runs Linux in production is testament to the stability of its SMP implementation.
And if you really want to look at some benchmarks at stability, have a look at this: