“Will the new Linux really perform in the same league as the big boys? To find out, I put the v2.6.0 kernel through several real-world performance tests, comparing its file server, database server, and Web server performance with a recent v2.4 series kernel, v2.4.23.” Read the performance article at InfoWorld.
Linux v2.6 Scales the Enterprise
About The Author
Eugenia Loli
Ex-programmer, ex-editor in chief at OSNews.com, now a visual artist/filmmaker.
Follow me on Twitter @EugeniaLoli
33 Comments
2004-02-01 6:45 pm
Anonymous
First of all, the fact that Slackware is not an extensively-documented Linux distribution, or that their (quite sparse) handbook doesn’t cover something, has nothing to do with the state of documentation in general. Slackware is easy to use if you already know what you’re doing. Anyone who doesn’t should be using Red Hat/Mandrake/Suse, or at least expect that they may run into some difficult problems.
Second: the quality of documentation about compiling kernels is going to have something to do with Linux getting ahead? That’s absurd. It’ll have no impact whatsoever on adoption outside of server rooms, where it has already been adopted to an impressive degree. Users who don’t want to have to deal with stuff like that use the hand-holding distributions I mentioned above.
Finally — what’s the deal? Do people really expect that modifying an operating system’s kernel will always be a cakewalk? And if so — why? This is something that most people will do never, a few people will do extremely rarely, and tinkerers might do all the time if they feel like it. Is it really that big an issue?
2004-02-01 7:15 pm
Anonymous
I try to load the page and get:
Welcome visitor!
Thank you for your interest in technology information from International Data Group, the leader in the field. We are eager to serve your information needs.
We regret that we can not satisfy this specific content request because it originates from a source that is not authorized to redistribute our material. Please access all of our rich store of technical knowledge directly by clicking on any of the following links:
<snip>
Thank you for respecting this necessary protection of our intellectual property.
Sincerely, IDG
———–
Can someone post the original text?
2004-02-01 7:39 pm
Anonymous
I have to agree with Rayiner Hashem on this. That is great that 2.6 is an improvement over 2.4 but how can these claims be made without benchmarking another operating system?
That is like me saying that Linux is a poor choice as a server because 2k3 server is faster than 2k.
PS. Anyone ever wonder why Windows server vs Linux benchmarks are nonexistent? The answer is that Linux has been killed in the few benchmarks that do exist. The Linux community flames them as being unfairly configured but never releases benchmark statistics of its own.
Maybe now that 2.6 is so much faster than 2.4 we can drop the emperors new cloths thing and actually benchmark these two products.
2004-02-01 8:09 pm
Anonymous
“then again I could be a f*cking idiot too?”
UltraSPARC IV worse than SPARC III? hahahahahahahahahahahahahahaha i bet your an insider
2004-02-01 8:19 pm
Anonymous
I bet you money Linux will smoke solaris on opteron hardware, unless if sun learned x86 well enough in such a short, short period of time. I think It’s just a matter of time that solaris will rise to be a great platform for x86 (32 and 64 bits) servers.
What servers would I run? Well right now I run opterons. Talk about a solid CPU… Would I run SPARCs? Yes… Would I run Linux 2.6? Yes… I am a linux worshipper? no…. Would I compare Linux to Win2k3? lol no…
I think Linux is already a major player. Linux needs better security, in my opinion thats whats bringing it down. It’s probably the most surperior OS for x86 hardware (other than FreeBSD) My question is that- linux for other platforms are relatively a new thing… How well will linux do on Itanium, PowerPC-64/32, SPARC?
If linux got its security under control and if 2.6 outperforms current stable builds will it unseat FreeBSD? I think It’s very likely. You have FreeBSD a huge open source project and Linux an even bigger Open source project with a viral license that can eat up BSD code if it wants. Linux has big time backers and big funding backing it… It will probably be the best x86-32 and x86-64 platform.. it might be just a matter of time.. But this is just a wild guess.. a big wild one.. I would do anything to see a benchmark of Linux 2.6 against FreeBSD 5.2 on a cheap cheap duron or athlon. I would!! That would be great!
2004-02-01 8:36 pm
Anonymous
Benchmarking Linux versus a commercial product may not be all that easy – many commercial software vendors have clauses in their licenses that forbid publication of such results without
company approval ( see link below)
And, if you read carefully through the so-called independent benchmarketing that we’ve seen to date, you’ll see that the playing field has never been close to level.
http://www.itworld.com/Man/2681/IWD010417opfoster/
What I’d like to see is a periodic public showdown, say 2 or 3
times a year, between the major players, maybe at Comdex?
Get the best minds from each camp to publicly tweak, tune and hammer on identical or equivalent hardware. Loser buys the beer.
2004-02-01 11:04 pm
Anonymous
Linux been running officially on non x86 hardware since version 2.0 (alpha) if not earlier.
as regards PPC64 well lot of work goes into this, i check the bitkeeper logs everyday and i regulary see ppc64 patches that have been applied by Linus. Then again given that the people who run that port are employed by IBM we shouldn’t really be surprised that it is well maintained.
2004-02-02 2:09 am
Anonymous
Very probably; I bet Linux isn’t that good with some workloads compared to ej: win 2k3. 2.6 has good SMP scalability, but solaris has been doing that for several years. It certainly will kill linux in bug servers, but I can bet Linux beats Solaris on smalls SMPs (ej: 2 cpus) and UPs. 2k3 probably has very good performance for several things, but I’d be very surprised if 2k3 can beat linux on some things like NUMA boxes (where the IBM people – one of the companies that sells NUMA machines – have put lots of efforts)
2004-02-02 2:10 am
Anonymous
“The Linux kernel has come a long way since Linus Torvalds’ announcement of v0.1 in 1991. The v2.6 kernel boasts many new features as well as major performance improvements over the v2.4 kernel and is poised to take Linux into the next stage of the game: true enterprise adoption.”
Like other posters have pointed out to support the above statement realworld tests need to be done against compairable commercial Unix systems.
It is good to see the linux kernel improve by leaps and bounds but if it is to be adopted in an enterprise environment it needs to perform as good if not better than the Solaris/HP-UX/AIX/ install base. Compairing it to Win2003 is pointless becasue linux already competes there on price alone.
2004-02-02 2:20 am
Anonymous
“Compairing it to Win2003 is pointless becasue linux already competes there on price alone.”
That’s not the right answer. Maybe for marketing people, but not for OSNews.
I bet the kernel developers would be very interested in looking at what loads linux sucks. There’re good reasons why you don’t see Linux people benchmarking against windows: People just don’t care. The one thing you can do with such benchmarks is “linux is X% better”, which isn’t really a exciting data. Kernel hackers benchmark against the previous release (aka 2.4). That’s the one thing it’s really useful for them: looking if they’re getting better or worse. Comparing with windows or other closed OSs doesn’t bring anything interesting except a possible “duh!”. Although personally I’d love to see such benchmarks, and I bet they’ll be interested in such benchmarks, but don’t asked them to do such benchmarks…
2004-02-02 2:40 am
Anonymous
Eugenia, I don’t know what sound card you have, but I compiled the 2.6 kernel with alsa support for my sound blaster live (EMU10K1) on Slack 9.1 and it works just fine.
That comment was really strange. How does testing Linux 2.6 against Linux 2.4 help you find out if “Linux reall perform[s] in the same league as the big boys?” I mean, okay, if you consider 2.4 one of the big boys…
v2.4 vs. v2.6? Wouldn’t a better comparision be between a popular flavour of unix, or against w2k3 server? Whats the point of comparing these to determine big boy status? All it determines is if 2.6 is better than 2.4.
That comment was really strange. How does testing Linux 2.6 against Linux 2.4 help you find out if “Linux reall perform[s] in the same league as the big boys?” I mean, okay, if you consider 2.4 one of the big boys…
I think he was also talking in relation to commercial UNIX’s out there. IMHO, one should wait till Solaris 10 and FreeBSD 5.x has stablised then conduct a review of them running on Opteron server.
Btw, it is going to be interesting to see what is released at the SUN Conference, 10 February 2004 this year. The UltraSPARC IV sounds very interesting, hopefully in six months they bring out a quad core CPU. There have been rumours also circulating about an 8way Opteron server being released at the end of this year, if so, it would be a good move on SUNs part.
WARNING: Kinda OT
I just had to bring this up. The documentation for recompiling a kernel are terrible. They are in short supply, missing crucial components(ex/the slackware handbook dosent mention modules_install), mostly geared towards RPM based distros(a good 90%+), and are out of date.
Linux needs more documenation if it evers whats to get ahead.
Very wow. For anyone who says Linux optimizes more for the desktop, eat it. You /can/ do both. 😀 Have you given 2.6 a whirl yet Eugenia?
>The documentation for recompiling a kernel are terrible.
Read here:
http://www.osnews.com/story.php?news_id=443
>Have you given 2.6 a whirl yet Eugenia?
I compiled it today and it initially seemd to work fine, except one critical problem: the hotplug stuff didn’t work on my slackware and so drivers for the network card, alsa, web cam etc etc didn’t load on startup. I had to modprobe them manually and in the case of alsa I would get a screech instead of sound… Anyone has a tip for that? It seems that the slackware scripts didn’t run, neither the modules.conf were taken into account over booting…
I found it kinda cool that the Opteron was faster in most of the tests. Go AMD!
> I think he was also talking in relation to commercial UNIX’s
> out there. IMHO, one should wait till Solaris 10 and FreeBSD
> 5.x has stablised then conduct a review of them running on
> Opteron server.
I agree that FreeBSD 5.x and Solaris will make interesting comparisons. You must realise though that FreeBSD 5.x is already stable enough for this kind of test despite the fact that it hasn’t been declared “STABLE”. It has had much more testing than Linux 2.6 and (in my opinion) has fewer issues left to iron out. Also, Solaris 9 and 10 are likely to perform in a similar way. 8 vs 9 was a different story. The switch from the M:N scheduler to the 1:1 made a few differences.
Linux 2.6, Solaris 9 and FreeBSD 5.2 can all be fairly compared today!
“I think he was also talking in relation to commercial UNIX’s out there. IMHO, one should wait till Solaris 10 and FreeBSD 5.x has stablised then conduct a review of them running on Opteron server.”
I agree that FreeBSD 5.x and Solaris will make interesting comparisons. You must realise though that FreeBSD 5.x is already stable enough for this kind of test despite the fact that it hasn’t been declared “STABLE”. It has had much more testing than Linux 2.6 and (in my opinion) has fewer issues left to iron out. Also, Solaris 9 and 10 are likely to perform in a similar way. 8 vs 9 was a different story. The switch from the M:N scheduler to the 1:1 made a few differences.
Linux 2.6, Solaris 9 and FreeBSD 5.2 can all be fairly compared today!”
The reason why I said later on was because the two need to settle down. Contra to fanboy opinions, FreeBSD 5.x is still not stable. As a desktop, probably but as a server being stress-tested, it still has some tender areas that are still in development, also, KSE would be required considering that it would be tested on a SMP configured Opteron along with certain Java related tests, meaning, it would be fairer for both sides to wait.
Solaris 10 also includes alot of new feature over 9. ZFS is one example which will replace UFS. There are numerous other features mentioned in an aceshardware.com article.
What is god’s name are you talking about?
Wait until the 2.6.2 release candidates are complete and then benchmark it. The 2.6.0 kernel is too raw.
At the end of March when the distributions ship 2.6, than match it up against Win3k, both on x86.
2.6 does not use modules.conf but modprobe.conf. use generate-modprobe.conf to convert.
Check out the scores!
But seriously was the kernel compiled properly or is the Itanium not suited to those tasks?
Or is it Linux’s history on x86 that gives it the boost?
I’d chalk it up to compilers. Itanium pushes more complexity into the compiler and its performance is more snesitive to such things then other archs. GCC is, from what I understand, no where near ICC on Itanium, despite gains on x86 to close the gap. I’d naturally assume Red Hat for Itanium is entirely compiled with GCC, and that its true performance is not evident based on that alone. Maybe with much better compilers the Itanium machine would be leading all those benchmarks. Hell, its likely I’d think.
Btw, it is going to be interesting to see what is released at the SUN Conference, 10 February 2004 this year. The UltraSPARC IV sounds very interesting, hopefully in six months they bring out a quad core CPU. There have been rumours also circulating about an 8way Opteron server being released at the end of this year, if so, it would be a good move on SUNs part.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
From what I heard the Ultra sparc IV is actually WORSE than USparc III, and the 8-way opteron… not sure if that is the one with 2 agp/3PCI-X/U320 SCSI/5.1 AUDIO and MUCH more… it is to be THE most feature packed MOBO!
but that may not be the same one.. it might have been 4-way
then again I could be a f*cking idiot too?
By Matthew Baulch
I agree that FreeBSD 5.x and Solaris will make interesting comparisons. You must realise though that FreeBSD 5.x is already stable enough for this kind of test despite the fact that it hasn’t been declared “STABLE”. It has had much more testing than Linux 2.6 and (in my opinion) has fewer issues left to iron out. Also, Solaris 9 and 10 are likely to perform in a similar way. 8 vs 9 was a different story. The switch from the M:N scheduler to the 1:1 made a few differences.
FreeBSD 5 is not that stable. It has recently switched to a different scheduler which is still plagued with problems. It has just the other day completely switched its threading library, neither of which have been default in any releases yet. Things like system call latency and context switching are way down from FreeBSD 4 (yes, with all debugging off).
FreeBSD 5 has definitely *not* had more testing than Linux 2.6. The Linux 2.6 userbase would vastly outnumber FreeBSD 5’s. Every day, labs at OSDL, IBM, SGI, Intel, Redhat, SuSE, Novell, etc rack up thousands and thousands of CPU hours every day of intensive testing. SGI is testing Linux 2.6 on 512 CPU supercomputers as we speak…
Linux 2.6, Solaris 9 and FreeBSD 5.2 can all be fairly compared today!
Definitely. However, one cannot say FreeBSD 5 is as stable or tested as linux 2.6.
Thanks Cristopher X.
I agree with what you are saying, it makes sense.
Even with gcc, at least kernel 2.6 gives it a performance boost.
“That said, the Opteron’s performance was outstanding; both the v2.4 and v2.6 kernels posted impressive results across all tests but most dramatically in the MySQL tests, showcasing the 64-bit support in v2.6.”
Wow, looks like AMD has a winner!
Cheers
WTF are you talking about? /usr/src/linux/README explains the process in all the detail you need, including the step:
” – If you configured any of the parts of the kernel as `modules’, you will have to do “make modules” followed by “make modules_install”.”
I’m not sure how much more documented you want it to be.
As was said before gcc can’t really compare to ICC when it comes to compiling on Itanium. Also HP’s compiler which i think is restricted to HP-UX & VMS has even better performance, given that HP designed McKinley (Itanium 2) this is hardly surprising tbh.
I forgot to mention SGI released an opensource version of their compilers with the MIPS backend ripped out, and Itanium backend added in, it’s the only opensource compiler on Itanium that supports Fortran 90/95
http://open64.sourceforge.net/
there’s no need to say, that the linux 2.6 kernel will suffer from tuning in the first releases. Look for example at Nick Piggin tests at http://www.kerneltrap.org/~npiggin/vm/3/“