A5/Linux bigadv - not using all cores?

The most demanding Projects are only available to a small percentage of very high-end servers.

Moderators: Site Moderators, PandeGroup

Re: A5/Linux bigadv - not using all cores?

Postby [Inpact]Terminou » Sat Apr 16, 2011 10:38 am

bollix47 wrote:I was seeing the same problem on Ubuntu 10.10 with fluctuating TPF and slower than windows. It seemed to be a scheduler problem so I updated my repos using linux-ck which uses the bfs scheduler.

I ran a regular smp project and am getting consistent time frames(i.e. all 24 cpus at or near 100% usage).

I will try running a bigadv tomorrow after the current one on the windows partition is complete.



That's the trick... Great thanks :D . Before i had TPF around 19/20 min (with kubunto 10.10 fresh install + all updates) instead of 15 in under seven for 6901 WU. By checking via htop tools it is better too.

I run kernel 2.6.35-28 . You can also cumulate thekrakkhen optimization and your TPF decrease of 30s :twisted:

I'm around 12min 47s.. I would like to see if we can improve with a kernel 2.6.38 with the same optimization. :eugeek:

This also improve for standard SMP WU.. I would like to know if an admin can make a topic for linux optimisation. Because it is tricky to troubleshoot a such issue.
EVGA SR2, 2* 5680,12 GO Corsair XMS3 Cas 9 2000MHz
MM Ext. Asc, Be Quiet 1200W P8
2 GWD GTX 260-896 GS 896 Mo (WC)
WD 600Go Sata III 10K, SSD Crucial C300 128Go (bigadv Kubuntu 10.10 + linux-ck + thekrakkhen)
All Watercooled
[Inpact]Terminou
 
Posts: 63
Joined: Sat Dec 05, 2009 6:24 pm
Location: Paris FRA

Re: A5/Linux bigadv - not using all cores?

Postby [Inpact]Terminou » Mon Jun 20, 2011 9:11 am

Totow can you have a look on that please ? You said nobody has to change the fah core ? Please have a look on the krakhen script and see if Stanford can officialese it or not internally.
[Inpact]Terminou
 
Posts: 63
Joined: Sat Dec 05, 2009 6:24 pm
Location: Paris FRA

Re: A5/Linux bigadv - not using all cores?

Postby bruce » Mon Jun 20, 2011 5:48 pm

Anyone can create a topic on Linux optimization -- probably best in one of the Linux-oriented forums. As long as it stays on the topic of optimizing FAH on Linux and doesn't get sidetracked into more general issues, there's no problem with that.

As far as getting Stanford to pass judgement on a specific 3rd party script or on specific bugs that a particular distro may have, that's not their job. They make sure that the science code produces good results. The Linux developers add features and may find bugs and correct them on a regular basis. This forum provides you with a method of sharing that information with each other and this topic is a perfect example.

Keep up the good work of reporting Linux features that work better/worse when running FAH. You guys are the experts as well as the judge and jury of what gets said.
bruce
 
Posts: 21936
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Previous

Return to SMP with bigadv

Who is online

Users browsing this forum: No registered users and 2 guests

cron