| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: The Clean Energy Project - Phase 2 Forum
Thread: Incorrect CPU usage time being reported and low credit granted [problem terminated] |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 33
|
|
| Author |
|
|
pirogue
Veteran Cruncher USA Joined: Dec 8, 2008 Post Count: 685 Status: Offline Project Badges: 
|
Now I'm trying 1 HCC and 1 CEP2. The utilization and elapsed/cpu time are still weird.
---------------------------------------- |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
CEP2 starts counting from zero time on each checkpoint in top, where BOINC of course continues cumulating. Compare task properties where it says "last checkpoint" and they're the same. CEP2 is the only one doing this AFAIK. Was conveyed to techs when first observed and confirmed.
----------------------------------------
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
fliegenpielz, yes top shows the percent per core up to 100% as does system monitor. I missed though where you get that first set where the Cpu0/1/2/3 is shown. Idle is I suppose the amount of time 'spare' per core from overall available for the nice processes. Rob: You can get the single-cpu-values by pressing '1' when in top. Yes, idle is just that, no process wants to have the cpu. Martin: NTFS is slow on linux, because it has to deal with a lot non-native security-stuff. cep2 permanently writes small chunks which causes this high load. |
||
|
|
sk..
Master Cruncher http://s17.rimg.info/ccb5d62bd3e856cc0d1df9b0ee2f7f6a.gif Joined: Mar 22, 2007 Post Count: 2324 Status: Offline Project Badges: 
|
OK, there is a problem writing to NTFS; its slow, but what if you are running Linux in RAM? Is this not faster?
----------------------------------------Noticed this oddity, Top -d10 CPU %CPU 2 . . 24.8 . . GPU Project 1 . . 19.7 . . WCG 0 . . 19.4 . . WCG 0 . . 17.8 . . WCG Everything else was at 0.0, except one task that was at 0.1. Note that CPU Core 0 is being used Twice! This just happens now and then, but it seems wrong. When just running 3 CEP2 tasks and one HCC task, 0.7 usr, 10.3 sys, 82.8 nic, 1.5 idle, I got the highest CPU usage I saw; 21, 20, 20, 19% CPU usage. It was usually a bit less, but still an improvement. With 4 CEP2 it is much less. No upload/download was in progress at the time. Boinc Manager was closed at the time, and nothing else open. The numbers seem to jump about too much, occasionally going down to 1% or so. All the way up to 13days now  but about 2.8 days in PV  This seems to be increasing, going by this thread [Edit 3 times, last edit by skgiven at Jul 9, 2010 4:37:59 PM] |
||
|
|
martin64
Senior Cruncher Germany Joined: May 11, 2009 Post Count: 445 Status: Offline Project Badges: 
|
Martin: NTFS is slow on linux, because it has to deal with a lot non-native security-stuff. cep2 permanently writes small chunks which causes this high load.  So what you are saying is that CEP2 ignores the setting of "only write to disk every ... seconds" and permanently writes stuff to the disk?  And if so, why does it only checkpoint once every one-point-something hours? Regards, Martin  [Edit 1 times, last edit by martin64 at Jul 9, 2010 11:17:56 AM] |
||
|
|
pirogue
Veteran Cruncher USA Joined: Dec 8, 2008 Post Count: 685 Status: Offline Project Badges:
|
 After doing a lot of searching on timing, clocks, etc., I stumbled across these kernel boot parameters. noapic acpi=off The result is in the screen capture. It's not 99%+ utilization, but it's better than 27.92% and the elapsed/cpu times are much closer. Now I'll actually get full(ish) time credit for these WUs. |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Hi piroque,
----------------------------------------At first thought there to be a typo in noapic because my LLL does not know it, but it was not. Great tracking and here's what my best friend, Google, found: Why do so many machines need "noapic"? and difference between noapic and acpi=off kernel parameters. There's lots more of course. Your sig is very 'current'... hope those struck will see it. Font size increase might help to spread the word a tad further. Thanks allot
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I guess I simply have to wait for the windows version in order to have a better performance.  What about just giving Linux one of it's native filesystems, ext3, ext4? Silly installs surely give silly performance. Only feel validations are tickling in a bit slowly. |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Martin: NTFS is slow on linux, because it has to deal with a lot non-native security-stuff. cep2 permanently writes small chunks which causes this high load. So what you are saying is that CEP2 ignores the setting of "only write to disk every ... seconds" and permanently writes stuff to the disk? And if so, why does it only checkpoint once every one-point-something hours? Regards, Martin Cannot confirm the constant writing of small chunks. I've got a fairly long WDS setting and BOINC is adhering to that for all sciences i.e. if checkpoints occur that are less than 10 minutes it's skipped. Theoretically that is only the case with CEP2 at the start but it's written anyhow per my client log, so I think the programmers have possibly not set that call for permission in CEP2 as in "dear core client, am I allowed now to write one?". A query by Jean is outstanding. If there is though a very large model that is not having sufficient space in RAM it starts swapping, which is what VM is designed for. BOINC I don't think has control over that part, just if the science app is set to ask for WDS permission. This web page explains a bit: http://plato.tp.ph.ic.ac.uk/~djw03/boinc-pyth...#boinc_time_to_checkpoint and http://boinc.berkeley.edu/trac/wiki/BasicApi#checkpointing As for the checkpoint bit, per previous discussions, CEP2 only checkpoints at 'job' end i.e. 16 times. Doing it more often is prohibitive due to the very large model size. As for the comments on file system, educational. Checked and mine is on ext4.. a sigh of relieve... just went with what the Lucid Lynx installer proposed, it breaking off it's own partition from the Windows C: drive and then formatting to what it now is.
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
martin64
Senior Cruncher Germany Joined: May 11, 2009 Post Count: 445 Status: Offline Project Badges:
|
What about just giving Linux one of it's native filesystems, ext3, ext4? Silly installs surely give silly performance. Definitely not. I don't need any Linux in my normal life, and buying a hard disk or fiddling around with the existing partition just to be 2 weeks ahead in WCG CEP2 is a lot more silly than what I did by booting from a memory stick and having boinc in the normal windows partition. It works fine with any other WCG project, only CEP2 puts such a high load on the mount.ntfs process. And only CEP2 refuses to run isolated on my larger memory stick, only produces errors there. It's not really silly performance, it's more that I find it silly that CEP2 continously reads or writes data to/from the disk. No swap, by the way, so the load cannot come from that. Regards, Martin |
||
|
|
|