benchmarking F@H

Moderators: Site Moderators, FAHC Science Team

alpha754293
Posts: 383
Joined: Sun Jan 18, 2009 1:13 am

Re: benchmarking F@H

Post by alpha754293 »

7im wrote:I recommend that you don't mix work units using a1 cores with a2 cores. They fold at different speeds and complicate the results. Use one or the other, or both seperately.
Well..since I can't freely swap them around, I'm just going to be storing one of the WUs.

Benchmarking with p2669 r16 c136 g56 on SLES 10 SP2 x64 (Linux release 2.6.16.60-0.21-smp #1 SMP Tue May 6 12:41:02 UTC 2008) (0)

*edit*
SLES 10 SP2 x64 results
(8*a2): p2669 r16 c136 g56 - 5945.81 PPD
(4*a2+8*a2): p2669 r16 c136 g56 - 5704.22 PPD (simulated HTT. One WU was used twice. F@H client invoked using "-smp 4" and "-smp 8" flags respectively). ~96% of single 8*a2 client PPD.

Conclusion: HTT should be DISABLED (based on the simulated HTT results).

Currently installing CentOS 5.2.

*edit*
Benchmarking with p2669 r16 c136 g56 on CentOS release 5.2 (Final) Kernel 2.6.18-92.el5 on an x86_64
Last edited by alpha754293 on Wed Jan 28, 2009 9:58 pm, edited 1 time in total.
alpha754293
Posts: 383
Joined: Sun Jan 18, 2009 1:13 am

Re: benchmarking F@H

Post by alpha754293 »

CentOS 5.2 x64 results:
(8*a2): p2669 r16 c136 g56 - 5945.81 PPD
(4*a2+4*a2): p2669 r16 c136 g56 - 6126.98 PPD

sidenote: load averages are quite a bit higher in CentOS. (7.7 vs 7.2 in SLES/SLED)

next up: Ubuntu 8.04 server x64
alpha754293
Posts: 383
Joined: Sun Jan 18, 2009 1:13 am

Re: benchmarking F@H

Post by alpha754293 »

Ubuntu 8.04.2 Linux 2.6.24-23-server

running into a litte bit of a hitch trying to get F@H running on it.

Code: Select all

# ls
fah6 mpiexec
# ./fah6
bash: ./fah6: No such file or directory
#
I have no idea what it's doing that. Any body who might know of something -- help at this point would be greatly appreciated otherwise, it's going to end up being dropped from the tests. (It really shouldn't be that difficult to install it and it get it up and running.) I've also simplified it for the time being. Once it's up and running again, I'll put the benchmarking WU back. But due to the unexpected difficulties encountered, I'm just trying to get it run to begin with.

*edit*
n/m. Got it. Someone should really write/include that for Ubuntu (at least for AMD64) that you need to run

Code: Select all

$ sudo apt-get install ia32-libs
in order to get F@H to run properly.

(not entirely sure if that applies to i386 installations.)

Also helps if you run this too to make your life simplier:

post-install, log in with the account that you created during the installation, and type this:

Code: Select all

$ sudo passwd root
so you can su to root and do everything you need to from there rather than having to always type "sudo".
Last edited by alpha754293 on Sun Jan 25, 2009 2:42 pm, edited 1 time in total.
bollix47
Posts: 2965
Joined: Sun Dec 02, 2007 5:04 am
Location: Canada

Re: benchmarking F@H

Post by bollix47 »

Have you installed the 32-bit libraries?

Code: Select all

sudo apt-get install ia32-libs
Also, have you made fah6 executable?

Code: Select all

chmod +x fah6
Image
alpha754293
Posts: 383
Joined: Sun Jan 18, 2009 1:13 am

Re: benchmarking F@H

Post by alpha754293 »

Ubuntu 8.04.2 server results:
(8*a2): p2669 r16 c136 g56 - 6331.60 PPD*

*NOTE: Running in text console only mode. No GUI. SLES and CentOS were running with GNOME.

next up: Ubuntu 8.10 server AMD64
alpha754293
Posts: 383
Joined: Sun Jan 18, 2009 1:13 am

Re: benchmarking F@H

Post by alpha754293 »

bollix47 wrote:Have you installed the 32-bit libraries?

Code: Select all

sudo apt-get install ia32-libs
Also, have you made fah6 executable?

Code: Select all

chmod +x fah6
yea I did. I found out after the fact though and I (always) run it as root in order to avoid permissions problems. (Not such a big deal if you're the only one on the system).
bollix47
Posts: 2965
Joined: Sun Dec 02, 2007 5:04 am
Location: Canada

Re: benchmarking F@H

Post by bollix47 »

Someone should really write/include that for Ubuntu (at least for AMD64) that you need to run

It's one of the first things mentioned in the Installation Guide:


http://folding.stanford.edu/English/LinSMPGuide
Image
alpha754293
Posts: 383
Joined: Sun Jan 18, 2009 1:13 am

Re: benchmarking F@H

Post by alpha754293 »

bollix47 wrote:
Someone should really write/include that for Ubuntu (at least for AMD64) that you need to run

It's one of the first things mentioned in the Installation Guide:


http://folding.stanford.edu/English/LinSMPGuide
oh..ha ha. my bad. *dawns dunce cap...again*.

(By the way...how can you tell if a distribution, ANY distribution of Linux is Debian or Redhat-based?)

While it says "some distributions", it might be nice to know specifically which ones you would need to and which ones you don't. So far, SLES, CentOS, and presumably RHEL, you don't have to. I'm guessing that with Ubuntu Desktop (i386 or AMD64), you don't have to either, but with server, you do (apparently).

Might I suggest that it be written something along this line instead:
"If you get a "No such file or directory" error and can't run it, you might have to get ia32-libs and install them using these commands "blah".

Only because that's kinda the tell-tale error that you'd get if you don't have it, which is what you'd need to do/run to get it.

That way, it'd also show up in search engines if people look for "fah6 No such file".

Might also help to put that it is also applicable to beta clients as well (so that when you run the search on it, you'd find it by keyword).
bollix47
Posts: 2965
Joined: Sun Dec 02, 2007 5:04 am
Location: Canada

Re: benchmarking F@H

Post by bollix47 »

By the way...how can you tell if a distribution, ANY distribution of Linux is Debian or Redhat-based?
Google is your friend. For example in the case of Debian-based distros:

http://www.debian.org/misc/children-distros
I'm guessing that with Ubuntu Desktop (i386 or AMD64), you don't have to either, but with server, you do (apparently).
Both AMD64 and server would require installing ia32-libs in Ubuntu. Some distros have it installed by default and use different names for the libraries containing the necessary libs as in RedHat for example.
The Linux SMP client is 32-bit, hence the requirement for the 32-bit libraries. This may change in the future. For example, see viewtopic.php?f=44&t=8041 for a discussion on the new 6.24 client which, at the moment is actually 64-bit.

Linux SMP is 64-bit only for the cores so i386 distros cannot run it.
Image
alpha754293
Posts: 383
Joined: Sun Jan 18, 2009 1:13 am

Re: benchmarking F@H

Post by alpha754293 »

Ubuntu 8.10 Linux 2.6.27-7-server AMD64 results:
(8*a2): p2669 r16 c136 g56 - 4411.91 PPD*

*NOTE: Running in text console only mode. No GUI. SLES and CentOS were running with GNOME.

And that concludes this testing.

WOW...I can't believe that Ubuntu 8.10 (default) is so bad. 30.3% slower. Although it's a known issue, but still....WOW. (Great for those who know how to download and update their kernels, but if you're totally new to it...you're stuck.)
alpha754293
Posts: 383
Joined: Sun Jan 18, 2009 1:13 am

Re: benchmarking F@H

Post by alpha754293 »

bollix47 wrote:
By the way...how can you tell if a distribution, ANY distribution of Linux is Debian or Redhat-based?
Google is your friend. For example in the case of Debian-based distros:

http://www.debian.org/misc/children-distros
I'm guessing that with Ubuntu Desktop (i386 or AMD64), you don't have to either, but with server, you do (apparently).
Both AMD64 and server would require installing ia32-libs in Ubuntu. Some distros have it installed by default and use different names for the libraries containing the necessary libs as in RedHat for example.
The Linux SMP client is 32-bit, hence the requirement for the 32-bit libraries. This may change in the future. For example, see viewtopic.php?f=44&t=8041 for a discussion on the new 6.24 client which, at the moment is actually 64-bit.

Linux SMP is 64-bit only for the cores so i386 distros cannot run it.
I'm reading this and watching Scrubs at the same time and I think I just pulled one of Dr. Cox's head shaking things.

That last part there, that has got to be one of the most confusing things ever. The clients are 32-bit, but it only runs on a 64-bit OS, because the cores are 64-bit?

I might switch to the 6.24 once I am done reinstalling my system since I just wrapped up doing the tests.
uncle_fungus
Site Admin
Posts: 1288
Joined: Fri Nov 30, 2007 9:37 am
Location: Oxfordshire, UK

Re: benchmarking F@H

Post by uncle_fungus »

Also helps if you run this too to make your life simplier:

post-install, log in with the account that you created during the installation, and type this:

Code: Select all

$ sudo passwd root
so you can su to root and do everything you need to from there rather than having to always type "sudo".
That's unnecessary, just use `sudo su` to enter a root shell without having to modify the root password on distros that use sudo.
alpha754293 wrote:That last part there, that has got to be one of the most confusing things ever. The clients are 32-bit, but it only runs on a 64-bit OS, because the cores are 64-bit?

I might switch to the 6.24 once I am done reinstalling my system since I just wrapped up doing the tests.
Yes, the clients are 32bit because they are "universal" clients which can launch CPU and SMP cores. 6.24 currently breaks this universality by being on 64bit at the moment though. SMP on Linux is 64bit only because the SMP core/MPI interaction doesn't work correctly on 32bit Linux at present.
Last edited by uncle_fungus on Sun Jan 25, 2009 4:23 pm, edited 1 time in total.
Reason: slow off the mark
smoking2000
Posts: 471
Joined: Mon Dec 03, 2007 6:20 am
Location: Amsterdam
Contact:

Re: benchmarking F@H

Post by smoking2000 »

uncle_fungus wrote:That's unnecessary, just use `sudo su` to enter a root shell without having to modify the root password on distros that use sudo.
Or just use `sudo -i` or `sudo -p` depending or your distro :)
uncle_fungus
Site Admin
Posts: 1288
Joined: Fri Nov 30, 2007 9:37 am
Location: Oxfordshire, UK

Re: benchmarking F@H

Post by uncle_fungus »

Fair point, I didn't know about the -i / -p options.
alpha754293
Posts: 383
Joined: Sun Jan 18, 2009 1:13 am

Re: benchmarking F@H

Post by alpha754293 »

uncle_fungus wrote:
Also helps if you run this too to make your life simplier:

post-install, log in with the account that you created during the installation, and type this:

Code: Select all

$ sudo passwd root
so you can su to root and do everything you need to from there rather than having to always type "sudo".
That's unnecessary, just use `sudo su` to enter a root shell without having to modify the root password on distros that use sudo.
alpha754293 wrote:That last part there, that has got to be one of the most confusing things ever. The clients are 32-bit, but it only runs on a 64-bit OS, because the cores are 64-bit?

I might switch to the 6.24 once I am done reinstalling my system since I just wrapped up doing the tests.
Yes, the clients are 32bit because they are "universal" clients which can launch CPU and SMP cores. 6.24 currently breaks this universality by being on 64bit at the moment though. SMP on Linux is 64bit only because the SMP core/MPI interaction doesn't work correctly on 32bit Linux at present.
Ahhh...that's a better explanation. What's interesting though is that NOW is the first time that a purely 64-bit release is made available.

I guess that they must have started the 32-bit development and kind fumbled along the way (stuff not working out as planned/expected) and then pushed over to 64-bit development and found that they got it to work before the 32-bit did. *shrug*

MPICH should work on 32-bit though. *shrug*

Hmm...interesting.
Post Reply