Benchmarks/Tests to run on Linux?

[Alphanos]

AFAIK, most of the oft-used benchmarks and tests are for windows. Where can I get some linux benchmarks to test the results of overclocking? I'll need to, for example, determine whether a certain speed/timings combo is better or worst than the last, and stuff like that.

On a slightly different topic, I've heard something about a memory testing program that you run overnight to confirm memory/system stability. Is this bootable? Where can I get it?

Thanks.

 

[Heisenberg]

There's a program called flops that I've used in the past for benchmarking. It's pretty minimal but it works. As for the memory tester, I think you're looking for memtest86. Do a quick google search and you'll find it.

 

[cleverhandle]

As far as I've seen, the *nix community doesn't put much faith or development time into synthetic benchmarks, apart maybe from SPEC which is commercial. There's an old benchmarking howto at TLDP, but it's pretty dated. And the few benchmarks I've scrounged up from there and elsewhere don't seem to have much solid data available for comparison. What seems more common is to run common tasks like a kernel compile, database operation, or GIMP transform and compare times. That will help you compare the effects of system configuration changes on a single machine, though not between machines. Many would argue that that's all benchmarks can ever be good for anyway.

 

[Nothinman]

For overclocking just pick something big to compile, if it segfaults you have a problem. There are other benchmarks like bonnie++ for hard disks and contest if you feel motivated.

memtest86 is a bootable floppy (based off a really old Linux kernel) that tests memory.

 

[cleverhandle]

Originally posted by: Nothinman
For overclocking just pick something big to compile, if it segfaults you have a problem.

I'll add that, in my experience, the kernel is not always tough enough to catch all problems. I recommend Mozilla as a good stress test, though it's a bit of a pain to set up the development environment for it. If possible, learn to use the source package tools for your system (dpkg-buildpackage or rpmbuild, for example) in order to streamline things. You don't want to near the end of a 90 compile and have it bomb because you forgot to set an environment variable somewhere.

 

[Nothinman]

apt-get build-dep mozilla
apt-get -b source mozilla

That's real hard =)

I guess timing a Gentoo installation could be considered a benchmark too heh.

 

[cleverhandle]

Originally posted by: Nothinman
apt-get build-dep mozilla
apt-get -b source mozilla

That's real hard =)

Will that work for repeated benchmarking without apt-get trying to re-download the dists each time? I would have thought to do...

apt-get build-dep mozilla
apt-get source mozilla
--Start here for repeats--
dpkg-source -x mozilla-blah-blah.dsc (not necessary the first time)
cd mozilla-blah-blah
time dpkg-buildpackage -rfakeroot -uc -b
cd ..
rm -rf mozilla-blah-blah

 

[Nothinman]

apt-get source kismet
Reading Package Lists... Done
Building Dependency Tree... Done
Need to get 685kB of source archives.
Get:1 http://http.us.debian.org sid/main kismet 2.8.1-4 (dsc) [664B]
Get:2 http://http.us.debian.org sid/main kismet 2.8.1-4 (tar) [678kB]
Get:3 http://http.us.debian.org sid/main kismet 2.8.1-4 (diff) [6219B]
Fetched 685kB in 4s (149kB/s)
dpkg-source: extracting kismet in kismet-2.8.1
warblade:/tmp/kismet$ apt-get source kismet
Reading Package Lists... Done
Building Dependency Tree... Done
Need to get 685kB of source archives.
Get:1 http://http.us.debian.org sid/main kismet 2.8.1-4 (dsc) [664B]
Get:2 http://http.us.debian.org sid/main kismet 2.8.1-4 (tar) [678kB]
Get:3 http://http.us.debian.org sid/main kismet 2.8.1-4 (diff) [6219B]
Fetched 3B in 0s (9B/s)
Skipping unpack of already unpacked source in kismet-2.8.1

Yes, it's smart enough not to redownload it.

 

[cleverhandle]

Cool. I wish I could be as smart as apt...

 

[jonmullen]

Originally posted by: Nothinman

I guess timing a Gentoo installation could be considered a benchmark too heh.

#time emerge kde
....gcc out put going on for DAYS

 

[Bremen]

Originally posted by: jonmullen

Originally posted by: Nothinman

I guess timing a Gentoo installation could be considered a benchmark too heh.

#time emerge kde
....gcc out put going on for DAYS

Depending on USE flags time can vary significantly. The first time I did it took about a day of computer time (PII 266, human had to sleep so computer sat idle at times). Second time I knew what I was doing more and set some USE flags... well installing the kernel sources required X, and lilo requires samba. It went something like lilo -> nasm, which has a doc flag -> ghostscript, which somehow goes to -> gimp -> gnome-libs ... -> samba. Oh well :0)

 

[Nothinman]

#time emerge kde
....gcc out put going on for DAYS

But my command doesn't actually install anything, it just compiles and makes an installable package.

 

[CCrunnernb]

emerge kde on my gentoo box took 12 hours...

thats with Dual 933 Xeons

then I hosed it all when i tried to partition another drive.. gah, gentoo was so fast

 

[CCrunnernb]

It would be nice to get ICC to compile things though, when you use ./configure CC=icc programs compile faster than hell *if* they will build with it...