Index  | Recent Threads  | Unanswered Threads  | Who's Active  | Guidelines  | Search
 

Quick Go »
No member browsing this thread
Thread Status: Active
Total posts in this thread: 11
Posts: 11   Pages: 2   [ 1 2 | Next Page ]
[ Jump to Last Post ]
Post new Thread
Author
Previous Thread This topic has been viewed 1964 times and has 10 replies Next Thread
BigMike
Cruncher
USA
Joined: Nov 14, 2005
Post Count: 13
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Page Faults

Thanks for the update! smile
While we're waiting, could someone look at the app (WinXP 32bit SP3)? It's generating a huge number (thousands per second) of page faults. And it seems to have nothing to do with available memory ... my system has at least 1GB free while CEP is running. No other grid app I run does this.

==Mike
----------------------------------------
Don't believe everything you think.
[Mar 24, 2009 3:58:21 PM]   Link   Report threatening or abusive post: please login first  Go to top 
Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline
Reply to this Post  Reply with Quote 
Re: Page Faults

http://wcg.wikia.com/wiki/Page_faults

edit: mere thousands per second is unfortunate, but usually not debilitating.
----------------------------------------
[Edit 1 times, last edit by Former Member at Mar 24, 2009 4:22:38 PM]
[Mar 24, 2009 4:20:09 PM]   Link   Report threatening or abusive post: please login first  Go to top 
David_L6
Senior Cruncher
USA
Joined: Aug 24, 2006
Post Count: 296
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: Page Faults

I guess that's just the nature of beast with CEP. I have a lot of page faults too. OS doesn't seem to matter (XP Home, XP Pro 32 bit, Vista 32 bit, and Vista 64 bit).
----------------------------------------

[Mar 24, 2009 7:41:41 PM]   Link   Report threatening or abusive post: please login first  Go to top 
BigMike
Cruncher
USA
Joined: Nov 14, 2005
Post Count: 13
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
Re: Page Faults

mere thousands per second is unfortunate, but usually not debilitating.


Well, on my system, we're talking about roughly 10,000 per second.

If we add together all the systems crunching CEP world-wide, we probably would have quite a few (hundreds? thousands?) CPU's worth of processing power doing nothing but servicing page faults.

Just because it's spread out across everyone's systems doesn't make it trivial. Still worth looking at. The grid is all about making small things large enough to matter.

==Mike
----------------------------------------
Don't believe everything you think.
[Mar 30, 2009 2:04:40 AM]   Link   Report threatening or abusive post: please login first  Go to top 
Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline
Reply to this Post  Reply with Quote 
Re: Page Faults

The techs decide where their efforts are best expended.

I saw recently a very useful rebuttal of that old saw "every little helps". Yes, every little helps - but only a little bit. In this case, your experience is exceptional, and you haven't even told us what the actual time cost is. Secondly, the problem only affects certain computer/task combinations, so the aggregate effect is even less significant.

Do we want to fix it? Of course! This issue is already on the list. But it will have to wait its turn behind more urgent things.
[Mar 30, 2009 9:30:37 AM]   Link   Report threatening or abusive post: please login first  Go to top 
Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline
Reply to this Post  Reply with Quote 
Re: Page Faults

I ran a quick study to see page faults on a linux system with a multi core cpu with the following results:
Note: I also abbreviated the terminal output of the sar command to reflect only the average page faults/second to keep the post more brief and legible.
System: AMD Phenom 9600 2300 MHz with Fedora 10

Page faults per second with BOINC client shut down (system background page faults reference):
[roger@starbase2 Documents]$ sudo sar -B 3 4
Linux 2.6.27.19-170.2.35.fc10.x86_64 (starbase2.command) 03/30/2009
09:17:59 AM fault/s Average: 16.37

Running CEP on 0 cores and HFCC on 4 cores simultaneously:
Linux 2.6.27.19-170.2.35.fc10.x86_64 (starbase2.command) 03/30/2009
09:25:18 AM fault/s Average: 53.04

Running CEP on 1 core and HFCC on 3 cores simultaneously:
Linux 2.6.27.19-170.2.35.fc10.x86_64 (starbase2.command) 03/30/2009
09:23:48 AM fault/s Average: 6525.48

Running CEP on 2 cores and HFCC on 2 cores simultaneously:
Linux 2.6.27.19-170.2.35.fc10.x86_64 (starbase2.command) 03/30/2009
09:16:32 AM fault/s Average: 13239.42

It would be interesting to quantify the total time to complete 4 CEP WU's when running simultaneously on 4 cores vs running the same WU's individually on the same single core but one would have to be able to run the WU's twice.
----------------------------------------
[Edit 1 times, last edit by Former Member at Mar 31, 2009 10:36:55 PM]
[Mar 30, 2009 5:32:30 PM]   Link   Report threatening or abusive post: please login first  Go to top 
Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline
Reply to this Post  Reply with Quote 
Re: Page Faults

mere thousands per second is unfortunate, but usually not debilitating.


Well, on my system, we're talking about roughly 10,000 per second.

If we add together all the systems crunching CEP world-wide, we probably would have quite a few (hundreds? thousands?) CPU's worth of processing power doing nothing but servicing page faults.

Just because it's spread out across everyone's systems doesn't make it trivial. Still worth looking at. The grid is all about making small things large enough to matter.

==Mike

I have to say he makes a very good point. While Didactylos said that it's on their list, I hope it's not too far behind, because it seems very important considering it could be costing so much cpu power. I mean, for me cep alternates between taking about 1k-2k every few seconds, and other projects like hfcc take maybe 8 every minute. It's defiantly important if you ask me.
[Mar 30, 2009 5:43:01 PM]   Link   Report threatening or abusive post: please login first  Go to top 
Sekerob
Ace Cruncher
Joined: Jul 24, 2005
Post Count: 20043
Status: Offline
Reply to this Post  Reply with Quote 
Re: Page Faults

Bluntly, it's economics. CEP takes about 8.6 CPU years for today, 271 CPU years are projected for the other sciences. The resources on WCG side are limited too, though it's costing their side 1.25 million Euro plus annual (actually I think it costs allot more, lots of hidden cost not passed within the large IBM organization), so choices which items to tackle first have to be made. One CEP job at a time does not seem to cause my quad to get saturated, so that's how I run them until a fix is found, probably a needle in a haystack stumbler.
----------------------------------------
WCG Global & Research > Make Proposal Help: Start Here!
Please help to make the Forums an enjoyable experience for All!
----------------------------------------
[Edit 1 times, last edit by Sekerob at Mar 30, 2009 6:21:30 PM]
[Mar 30, 2009 6:16:42 PM]   Link   Report threatening or abusive post: please login first  Go to top 
Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline
Reply to this Post  Reply with Quote 
Re: Page Faults

And still nobody has quantified what it actually costs in terms of kernel time.
[Mar 30, 2009 6:32:51 PM]   Link   Report threatening or abusive post: please login first  Go to top 
Sekerob
Ace Cruncher
Joined: Jul 24, 2005
Post Count: 20043
Status: Offline
Reply to this Post  Reply with Quote 
Re: Page Faults

Darn, just see 2 CEP running on the quad: 1 9:47 hours with 14 minutes Kernel time, the other 6:56 hours with 10 minutes.
----------------------------------------
WCG Global & Research > Make Proposal Help: Start Here!
Please help to make the Forums an enjoyable experience for All!
[Mar 30, 2009 6:38:17 PM]   Link   Report threatening or abusive post: please login first  Go to top 
Posts: 11   Pages: 2   [ 1 2 | Next Page ]
[ Jump to Last Post ]
Post new Thread