| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: Help Conquer Cancer
Thread: HCC 5.07 and page faults |
| No member browsing this thread |
|
Thread Status: Locked Total posts in this thread: 16
|
|
| Author |
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Q6600 on stock speed 2.4ghz, FSB 1065mhz, 4 HCC jobs concurrent with 2gb RAM 667mhz, fixed VM, Vista HP. The PF errors are between 4 and 175k and barely moving. The Delta is continuously zero, watching it for minutes on end. According the message log the jobs checkpointed multiple times during the period of observation. All the jobs logged for 5.15 are in the 4:06 to 4:25 hour range on this machine.
----------------------------------------Interesting Lawrence. The CPU-Z utility reports 4x32kb L1 and 2x4096 L2 cache. As for the related thread on claims/credit, here's how it looks from my perspective. X0000042490704200412132149_ 1-- xxxxxxxxxx Valid 11/23/2007 02:44:02 11/26/2007 11:41:57 4.44 68.0 / 68.0 X0000042490643200412132150_ 0-- xxxxxxxxxx Valid 11/23/2007 02:42:24 11/26/2007 10:17:02 4.01 61.7 / 70.9 X0000038190932200410011730_ 0-- xxxxxxxxxx Valid 11/22/2007 05:01:50 11/26/2007 10:03:13 4.16 64.1 / 62.4 X0000038180082200410011718_ 1-- xxxxxxxxxx Valid 11/22/2007 02:59:31 11/26/2007 09:22:53 4.17 64.2 / 64.2 X0000038180174200410011716_ 1-- xxxxxxxxxx Valid 11/22/2007 03:01:11 11/26/2007 09:22:53 4.15 63.9 / 70.4 X0000038180217200410011715_ 1-- xxxxxxxxxx Valid 11/22/2007 03:02:53 11/26/2007 09:22:53 4.12 63.4 / 60.6 X0000038861378200409221652_ 0-- xxxxxxxxxx Pending Validation 11/21/2007 20:39:27 11/23/2007 19:51:30 4.09 63.1 / 0.0 X0000039760062200409281037_ 1-- xxxxxxxxxx Valid 11/21/2007 23:15:51 11/23/2007 19:51:30 4.08 62.9 / 65.9 The top one is one of these anomalous claims X0000042490704200412132149_ 0-- Valid 11/23/2007 02:44:02 11/24/2007 10:29:14 10.48 126.4 / 68.0 < PF victim? X0000042490704200412132149_ 1-- Valid 11/23/2007 02:44:02 11/26/2007 11:41:57 4.44 68.0 / 68.0 < Moi The top in the above quorum based on hourly claim should have done it in about 5.6 hours.
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at Nov 26, 2007 11:57:19 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Page faults are usually caused by 'locality of reference'. If the algorithm does not reference nearby array values, then the cache will not have the requested information. This is probably a necessity of the algorithm, in which case the programmers cannot do anything about it. Both Athlon X2 3800+ and Athlon 2500+(Barton) have 512 L2 cache (per core) but PF Delta value differ |
||
|
|
Sgt.Joe
Ace Cruncher USA Joined: Jul 4, 2006 Post Count: 7846 Status: Offline Project Badges: 
|
I checked the page fault delta and it was around 17,500. Must be something about the way this application makes use of some piece of the hardware. If others with more memory are also experiencing this behavior it is probably not a constrained memory, but Didactylos may be onto something. I am purely guessing but think the thrashing is probably causing at least a 50% slowdown. On the bright side at least the result was valid.
----------------------------------------Cheers
Sgt. Joe
*Minnesota Crunchers* |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hello s060319, Cannot they set any good programmers on the job. Page faults are usually caused by 'locality of reference'. If the algorithm does not reference nearby array values, then the cache will not have the requested information. This is probably a necessity of the algorithm, in which case the programmers cannot do anything about it. Lawrence Maybe they can do something about it. They can change the order of the program. As you can see at the bottom of this page: http://www.worldcommunitygrid.org/projects_sh...Hcc1Faq.do?shortName=hcc1 The first step in making the three gray pictures in there they use the original picture. They work with the same pixels to calculate the (co)variances of the pixels. If the program loads the pictures in the cache it can also calculate the middle and the lowest picture. This is a way to reduce the pagefaults. I don't know if they already use this method. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I checked my PFs based on this - the HCC tasks are regularly coming in over 1 billion (1,000,000,000+) page faults for each completed job. (2 processor, 2 cores each, 3ghz xeons) That can't be helping the performance. Still, all the results seem to end up valid.
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Okay, thank you all for your help.
We know about the problem, the techs know about the problem. Right now they are working on more urgent problems. So, thank you all for your help. Topic closed. |
||
|
|
|