This article will show you how to benchmark operating system performance using MySQL on these OSes so you can find out for yourself if you’re missing out. While this may not necessarily be indicative of overall system performance or overall database application performance, it will tell you specifically how well MySQL performs on your platform.
Using MySQL to benchmark OS performance
About The Author
Eugenia Loli
Ex-programmer, ex-editor in chief at OSNews.com, now a visual artist/filmmaker.
Follow me on Twitter @EugeniaLoli
<quote>
This article will show you how to benchmark operating system performance using MySQL … While this may not necessarily be indicative of overall system performance or overall database application performance, it will tell you specifically how well MySQL performs on your platform.
</quote>
So this shows you how to benchmark your OS using MySQL as the benchmarking tool, OH no – wait, it shows you how to benchmark MySQL on your OS.
When reading that statement, does anyone else scratch their head and say “What?”.
I have to agree with you, this benchmark cannot be used to measure the speed of the OS accurately. It would depend greatly on how optimized MySQL was for your system.
What a let down. Scroll to the “conclusion” and it sez better-luck-next-time? “I’ll post the results in my next article?”
And here I was all psyched and ready for a massive flame-fest on how the bench marks can’t possibly be accurate because of reason X and how this research was sponsored by an Evil Corporation(TM) and how he should have compiled with lib_hyperextremeparallelthreading4.5.2.45 !
<glumly-removes-flame-suit />
I’ve heard MySql has spent considerable time optimizing it’s preformance on Linux and Solaris. To the point of working with the Linux and Solaris developers threading issues.
So, we shall see, rumor or fact…
If Linux beats out all the operating systems, I whole heartedly and completly support the authors noble endeavor.
Otherwise, I challenge his methods and inate bias’s.
bias’s?
http://angryflower.com/bobsqu.gif
I figured it would be bias’s because they belong to the author; hence the possesive. If this is wrong: My Bad.
I think the way you used bias the correct spelling is simply “bias.” Would you say: “I challenge his methods and hairy foot’s?” That’d be not only silly, but all wrong. It should be: “I challenge his methods and hairy foot.” See? No possessive needed there. Bias works the same way. Besides, you already showed the possessiveness via the usage of “his” and need not duplicate the possessive indication later with an apostrophe. Otherwise, you’d end up with sentences like this: “I heard his nose’s is broken.” That just sounds wrong, doesn’t it? 🙂
Yes Mr. BigZaphod
I will write that on the board 100 times.
No way are you a native english speaker, you know way to much about the english language.
Here are my predictions for that benchmark :
First place (a tie) : Linux 2.6; NetBSD 2.0
Second place (another tie) : FreeBSD 4.11; Solaris 10
Third place : Linux 2.4
Fourth place : FreeBSD 5.3
Disqualified : OpenBSD 3.6
OpenBSD 3.6 was thrown out of the competition because it was drunk threading : it saw one CPU instead of two. Also, it wasted time encrypting the filesystem, only to die after a kernel panic. The author wished he had used the latest snapshot of DragonFly BSD.
Just kidding 🙂
It’s *possible* Adam meant to use “bias”. And made a mistake about affixing the possessive. But most such mistakes are the result of transcribing the sounds one has in mind into incorrect spelling. That is, the sounds are correct, the spelling is wrong.
Adam probably meant “biases”, the plural, as indicated by the preceding “methods”. Just a hunch.
(After all that, I really like Adam’s comment, and that of Mr. Glumly Removing Flame-suit. Have to wait for our pyromania to be satisfied, vicariously of course.)
My predictions
Linux 2.6
Solaris 10
FreeBSD 5.3
Linux 2.4
FreeBSD 4.11
NetBSD 2
OpenBSD 3.6
Linux and Solaris I predict will be close, Solaris may take it though I have a affinity for Linux systems, so I give it favor. I also predict FreeBSD 5.3 and Linux 2.4 will be close.
I want to see more predictions. Just for fun, nothing serious or anything, I just think it’s interesting to see people’s guesses.
Solaris 10
Linux 2.6
NetBSD 2
OpenBSD 3.6
Linux 2.4
FreeBSD 5.3
FreeBSD 4.11
Don’t forget that he’ll be running on a dual CPU system. Although not going over the network (which tends to be very kernel intensive), OS scalability may start to show some influence.
Linux 2.6
Solaris
Linux 2.4
NetBSD
FreeBSD 4
OpenBSD
FreeBSD 5
He’s testing both single cpu and smp modes.
mySQL naturally does better one some OS’s than others. For example, on many FreeBSDs you’d need to install some extra software for mySQL to run threads fast.
mySQL naturally does better one some OS’s than others. For example, on many FreeBSDs you’d need to install some extra software for mySQL to run threads fast.
A MySQL workload will present itself to the kernel as a profile of system calls and IO (interrupts, etc). Now considering it won’t be a network test, and the same hardware will be running all tests, that takes the IO side out of the equation. So what you are left with is a comparison of the efficiency at which each operating system can complete a set of system calls.
If you installed some extra software for MySQL for FreeBSD, then *that* would be unfair.
1. NetBSD 2.0
2. FreeBSD 5.3
3. OpenBSD 3.6
4. Solaris 10
5. Linux 2.6
6. FreeBSD 4.11
7. Linux 2.4
Congrats NetBSD Team !!!
Yes I am certain that OpenBSD will trounce Linux 2.6… makes perfect sense…
reiserfs needs kernel_lock() protection due to bad coding, it’s a fs with big latencies in no way suitable for smp.
it’s a really bad and uninformed choice of a linux journilling fs. Anyway which bsd does support journalling?
out of fairness ext2 would be a much better choice, or xfs/ext3 if you really want an journalling fs.
These newsforge articles are so shallow, yes mod me down, but it has to be said.
For single cpu
Linux 2.6
Solaris 10
NetBSD 2
Linux 2.4
FreeBSD 4.11
FreeBSD 5.3
OpenBSD 3.6
I predict Linux 2.6 and Solaris 10 will be close, and NetBSD 2 and Linux 2.4 will be close — the resaults may be interchangable.
The results are up:
http://software.newsforge.com/article.pl?sid=04/12/27/1243207&from=…
Looks like Linux 2.6 and 2.4 practically sweeped the resaults. I expected Solaris to win a few, and I did not expect NetBSD to beat Linux for single CPU for the first benchmark. I did not expect 2.4 would do so well enlight of the fefe scalability benchmarks. http://bulk.fefe.de/scalability/
As it turns out the Solaris 10 benchmarks that were suppose to dominate Linux, were probably hyped and wrong. Likewise it turns out that the FreeBSD 5.3 smpng infrastructure and their semaphore-like approach did not do that much, albeit, it is not complete and their VFS is still tied to their giant lock.
It seems Linux is still on top for performance issues. I am looking foward to future releases of Linux and hope FreeBSD 6 is not a dud.
As it turns out the Solaris 10 benchmarks that were suppose to dominate Linux, were probably hyped and wrong. Likewise it turns out that the FreeBSD 5.3 smpng infrastructure and their semaphore-like approach did not do that much, albeit, it is not complete and their VFS is still tied to their giant lock.
But regardless of that, it is good to see Linux 2.6 demolishing all in its path! The single threaded performance of Linux 2.4 and more scalable than Solaris (the scalability ratio is the same as solaris, but the overall results are higher, and thus more impressive).
NetBSD had one impressive result, but rolls over and dies when the going gets tough (Sysbench 10M rows). Their CPU scalability is horrible though.
It is a relief to see FreeBSD 5 is somewhat scalable, but the lackluster single processor results still confirm it isn’t a good option for performance critical applications.
And Solaris 10, quite apart from being trounced by Linux 2.6, couldn’t even beat Linux 2.4. For shame.
The response from the FreeBSD camp?
http://marc.theaimsgroup.com/?l=freebsd-current&m=110797925430587&w…
Linux can’t possibly be faster than us so they must be cheating.
This is exactly the kind of thinking that has been derailing the FreeBSD project for the past 5 years or so.
FYI, NetBSD is going to have to be redone. The author missed an undocumented procedure that actually allows the 2nd processor to do something.
Yes, undocumented. You can find it in the mailing lists with the topic “documentation problems”.
The actual param is: PTHREAD_CONCURRENCY
Also, the Gentoo setup was probably a pretty standard setup meaning no major optimizations out of the box but more to the point, PIE and SSP hardened C compiler. The Linux systems may have done better with a true Gentoo guru.