| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: BOINC Agent Support
Thread: WCG client throughput and optimisation |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 15
|
|
| Author |
|
|
Trotador
Senior Cruncher Joined: Mar 26, 2009 Post Count: 154 Status: Offline Project Badges: 
|
I have looked at my stats per PC and I've noticed something that probably is easy to explain but I would like to know other opinions.
----------------------------------------I have two quads fully dedicated to WCG (no use for anything else), only other task running is a folding@home GPU client that consumes little to nothing CPU power. I was expecting to find that each PC was dedicating almost 4 days to WCG. To my surprise it is actually much lower a very few days it reaches over 3 days and 12 hours. Is it normal that overhead because of OS and whatever other background tasks? What are your experiences and data? These are XP32bits with 2 GB RAM machines. Thanks for sharing Edit; an example from one of them 2nd edit: this one has 1 GB RAM 20/09/09 0:003:15:44:56 11.506 19 19/09/09 0:003:04:09:33 10.211 17 18/09/09 0:003:02:01:28 9.909 16 17/09/09 0:002:21:57:55 9.338 16 16/09/09 0:003:10:27:42 10.911 18 15/09/09 0:002:05:16:18 7.044 12 14/09/09 0:003:13:07:04 11.047 19 13/09/09 0:004:07:21:40 13.411 22 12/09/09 0:004:00:24:36 12.441 21 11/09/09 0:004:09:54:00 13.720 23 10/09/09 0:003:06:15:46 10.106 17 2nd edit: this one has 2 GB RAM, it is a better average 20/09/09 0:003:15:49:34 13.012 19 19/09/09 0:004:00:20:38 14.200 20 18/09/09 0:003:22:06:35 14.035 20 17/09/09 0:003:19:21:05 13.527 20 16/09/09 0:003:17:12:58 13.244 19  [Edit 2 times, last edit by Trotador at Sep 21, 2009 8:01:37 PM] |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges: 
|
Is it normal that overhead because of OS and whatever other background tasks? I think it is not normal.Use TaskManager sorted on Total CPU Time to find the culprit(s) and to check that the close-to-nothing of the GPU task is really close-to-nothing. For the 1 GB machine maybe there could be some memory swapping depending on the projects you run, but for the 2 GB machine RAM should be all fine and, still, it loses almost one day over a period of 5. |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges: 
|
Well, since it seems you're also running other BOINC-projects, the easy question 1st. is, you're not by chance also running one or more active projects on either of the affected computers? (I don't count projects like SIMAP that haven't had work for the last weeks).
----------------------------------------If no... What is your on_frac, active_frac and cpu_efficiency? You can find these nearly at the start of client_state.xml (located in BOINC data-directory, can open with example internet explorer), or, if attached to a BOINC-project except WCG, hit "update" on one of the projects except WCG, and afterwards look on this projects computer info. On computer-info it will look something like this: % of time BOINC client is running 99.7373 %  "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." [Edit 1 times, last edit by Ingleside at Sep 21, 2009 11:24:11 PM] |
||
|
|
Trotador
Senior Cruncher Joined: Mar 26, 2009 Post Count: 154 Status: Offline Project Badges:
|
Thanks for the answers, no other BOIN project has been executed during the days I showed the data. The task manager does not show any task taking much time but think could be there anyway...wait the SO inactive task is over 130 hours in the 1GB RAM machine, so it seems clear that the 1 GB RAM is limiting that one, I see over 1 GB memory used in the task manager, so swaping can be definitively an issue here, it is running four IADS
----------------------------------------the client_state.xml shows: - <time_stats> <on_frac>0.850971</on_frac> <connected_frac>0.999931</connected_frac> <active_frac>0.999797</active_frac> <cpu_efficiency>0.984727</cpu_efficiency> <last_update>1253595419.037499</last_update> </time_stats> The 2 GB RAM machine shows: - <time_stats> <on_frac>0.999800</on_frac> <connected_frac>0.999969</connected_frac> <active_frac>0.999914</active_frac> <cpu_efficiency>0.984167</cpu_efficiency> <last_update>1253595602.031250</last_update> </time_stats> time to purchase nore RAM to begin with. |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges:
|
Fine, you are progressing. While waiting for its additional RAM you might switch the 1 GB machine to HCMD2 or Rice which have the lowest memory footprint of our projects.
----------------------------------------Still there are those 20 % lost in the 2 GB machine... If you don't find naughty background processes anywhere you might check the so called "optimization" features of Windows which slow machines all day long for possibly saving you a few seconds once in a while. The one I am thinking of right now is the Indexing Service which scans files for speeding up file searching. In my opinion this is definitely useless in an unattended machine (and possibly in most attended ones too  ).Good luck. Jean. |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
The indexing service on Vista SP2 level and XP-SP3 actually improved substantially, once they are fully indexed, but I do exclude zones such as the BOINC data area. It's now always on and find the x-devices backup synching to be at outstanding speed. Task manager shows them taking very little time these days. 6:34 minutes since last boot, 5 days ago on a quite actively used quad.
----------------------------------------
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges:
|
Thanks for the answers, no other BOIN project has been executed during the days I showed the data. The task manager does not show any task taking much time but think could be there anyway...wait the SO inactive task is over 130 hours in the 1GB RAM machine, so it seems clear that the 1 GB RAM is limiting that one, I see over 1 GB memory used in the task manager, so swaping can be definitively an issue here, it is running four IADS the client_state.xml shows: - <time_stats> <on_frac>0.850971</on_frac> <connected_frac>0.999931</connected_frac> <active_frac>0.999797</active_frac> <cpu_efficiency>0.984727</cpu_efficiency> <last_update>1253595419.037499</last_update> </time_stats> According to <on_frac>, this computer is turned-off, or BOINC isn't running, for roughly 4 real hours/day. So, if computer haven't been totally off for some days and when turned-on again and afterwards been running 24/7, being a quad it's been losing roughly 15 cpu-hours/day... Or, in other words, based on your <on_frac> your quad is only running 3.4 days per day, something that seems to be more or less in line with your results... The <cpu_efficiency> is excellent, losing only 1.6% is basically that you'll get doing some web-browsing and so on, so atleast from this GPU-folding has had negligible effects, and there's been no other process stealing cpu-time. Installing more memory would maybe be an improvement, but not when it looks like the culprit is that BOINC has been turned-off for some hours per day... "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges:
|
Ingleside,
----------------------------------------I often wondered: on how many days is the on_frac ratio computed? Or if you prefer, after how many days of full activity is a vacation period completely out of the picture? I have the feeling that it takes a while. |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges:
|
Ingleside, I often wondered: on how many days is the on_frac ratio computed? Or if you prefer, after how many days of full activity is a vacation period completely out of the picture? I have the feeling that it takes a while. BOINC uses exponential weighting on the time_stats, so it should be something similar to RAC. Now, not sure if it can be called "half-life", but atleast they're using 10 days on the time_stats, except cpu_efficiency used 1 day instead. So, if not mistaken, if starts with zero time_stats, after running 24/7 for 1 day cpu_efficiency would have increased to 0.5, and after 10 days the other time_stats is 0.5. BTW, then it comes to very long outages, like for holidays and so on, if the outage is over 14 days it's ignored in v6.10.4 and later clients. "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
Trotador
Senior Cruncher Joined: Mar 26, 2009 Post Count: 154 Status: Offline Project Badges:
|
Ok, upgraded to 2 GB RAM (btw DDR2 RAM is dirty cheap these days), index server deactivated in all machines, let's see next days.
----------------------------------------The previous 1 GB RAM machine has been switched off around one week or ten days before this run, but from then on 24/7. Now, I will maybe reach emeral badge for IADS before than foreseen before  .Thanks for the help, |
||
|
|
|