Message boards :
Number crunching :
AstroPulse v6.02: 16.59 credits for 91.7 hours' processing?!
Message board moderation
Author | Message |
---|---|
Nigel Garvey Send message Joined: 18 May 99 Posts: 39 Credit: 2,228,452 RAC: 3 |
|
Khangollo Send message Joined: 1 Aug 00 Posts: 245 Credit: 36,410,524 RAC: 0 |
This always happens when you're running anonymous platform and your wingman is running nVidia/ATI Astropulse GPU application (CreditNew - the scoring system s@h uses doesn't like it at all). Weird enough, you're running stock application, so this shouldn't happen. But it did. My suggestion is don't run astropulse on slow computers. Stupid CreditNew will "rip you off" often, so at least let it rip you off on shorter workunits. |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
... Stock application with less than 10 "completed" doesn't override anonymous platform. Joe |
Nigel Garvey Send message Joined: 18 May 99 Posts: 39 Credit: 2,228,452 RAC: 3 |
Thanks for the reply. I wouldn't yet call my 2.53 GHz Intel Mac a "slow" computer, although it is a couple of years old now. However, the AstroPulse application it downloaded has "powerpc" in its name (I don't know why SETI@home sends PowerPC apps to Intel machines!) so maybe having to run the task via Rosetta is what made it take so long. My slower, PowerPC machine has completed an AstroPulse task in just over two thirds the time and is still waiting for the wingman to report. NG |
Fred J. Verster Send message Joined: 21 Apr 04 Posts: 3252 Credit: 31,903,643 RAC: 0 |
... The first AstroPulse tasks, came with an estimate of >200 hours, but this quickly changed. So far, I've 1 AstroPulse WU completed, checked and made a cannonical result, This I7-2600 + ATI host. Percentage of blanking shows in the CPU use/time. CPU v.s ATI GPU. . |
arkayn Send message Joined: 14 May 99 Posts: 4438 Credit: 55,006,323 RAC: 0 |
Thanks for the reply. I wouldn't yet call my 2.53 GHz Intel Mac a "slow" computer, although it is a couple of years old now. However, the AstroPulse application it downloaded has "powerpc" in its name (I don't know why SETI@home sends PowerPC apps to Intel machines!) so maybe having to run the task via Rosetta is what made it take so long. My slower, PowerPC machine has completed an AstroPulse task in just over two thirds the time and is still waiting for the wingman to report. I sent off a report to Eric about the wrong app issue. |
Eric Korpela Send message Joined: 3 Apr 99 Posts: 1382 Credit: 54,506,847 RAC: 60 |
David considers this a feature of the way the scheduler works not a bug, and he didn't like my suggestion of numerical plaform priorities or estimated relative speeds for multiple platforms. There's a way to turn this feature of, but not without causing more problems than it solves on other platforms. To make a really long story long, the BOINC 6.x Mac client for intel reports that it is capable of running both Intel and PPC apps (which may or may not be true). The BOINC 7.x Mac client only reports that it can run Intel apps because there is no BOINC 7.x Mac PPC client. When the scheduler gets a request from a BOINC 6.x Mac intel it calculates the relative speed of the intel and ppc and (most of the time) concludes that the intel version is 3x the speed of the PPC version, and so it sends the intel version. But in order to reach that conclusion it occasionally has to send out the PPC version so it can see how fast it is. It does that by adding a random value to the speed of the PPC verion and one out of every few thousand results it will send out the PPC version as a test. I'm in the process of testing a modification to that method on beta which should reduce the number of PPC results sent out to Intel Macs once it becomes clear that the PPC version is slower. This thing will really become important as projects have more VM, emulated, or java apps in addition to native apps. Till then, if you don't want the PPC work on your Intel machine, about all I can suggest is upgrading to BOINC 7 if you can. Sorry I wont be able to get a better solution deployed in a rapid fashion. Eric @SETIEric@qoto.org (Mastodon) |
Eric Korpela Send message Joined: 3 Apr 99 Posts: 1382 Credit: 54,506,847 RAC: 60 |
And, unfortunately the low credit granted was due to a host using a new anonymous platform that hadn't worked through its first 10 results to get calibrated and thus was in the "approximate" credit state. For two results in that state, the low claim gets rewarded, so you got credit for 3400 seconds of his CPU. Had you or your wingman been calicrated "normal" credit, it probably would have been about 700 credit. Another thing I need to fix if I can figure out how and find the time. @SETIEric@qoto.org (Mastodon) |
perryjay Send message Joined: 20 Aug 02 Posts: 3377 Credit: 20,676,751 RAC: 0 |
Remind me not to get teamed with another ATI GPU when I do APs... http://setiathome.berkeley.edu/workunit.php?wuid=982683654 19.66 credits just don't cut it. :( PROUD MEMBER OF Team Starfire World BOINC |
Eric Korpela Send message Joined: 3 Apr 99 Posts: 1382 Credit: 54,506,847 RAC: 60 |
I'm looking right now at the logs to figure out how that one happened. Both of you had enough normal results that you shouldn't have been requesting such a tiny amount of credit. Well here's the log segment. 2012-05-03 17:22:10.9158 [credit] [RESULT#-1872162716] raw credit: 79.22 (3422.29 sec, 10.00 est GFLOPS) 2012-05-03 17:22:10.9171 [credit] [RESULT#-1872162716] anon platform, scaling by 0.248142 (0.17/0.67) 2012-05-03 17:22:10.9184 [credit] [RESULT#-1872162716] anon platform, returning 19.66 2012-05-03 17:22:10.9199 [credit] [RESULT#-1872162716] updating HAV PFC 0.02 et 1.87929e-12 turnaround 48582 2012-05-03 17:22:10.9214 [credit] [RESULT#-1872162716] get_pfc() returns credit 19.6577 mode approx 2012-05-03 17:22:10.9227 [credit] [RESULT#-1872162715] raw credit: 196.47 (8487.64 sec, 10.00 est GFLOPS) 2012-05-03 17:22:10.9240 [credit] [RESULT#-1872162715] anon platform, scaling by 3.75974 (0.17/0.04) 2012-05-03 17:22:10.9253 [credit] [RESULT#-1872162715] anon platform, returning 738.69 2012-05-03 17:22:10.9267 [credit] [RESULT#-1872162715] updating HAV PFC 0.05 et 4.66085e-12 turnaround 222419 2012-05-03 17:22:10.9280 [credit] [RESULT#-1872162715] get_pfc() returns credit 738.688 mode approx 2012-05-03 17:22:10.9297 [credit] [WU#982683654] assign_credit_set: credit 19.6577 2012-05-03 17:22:10.9318 [debug] [HAV#12000004] consecutive valid now 152 2012-05-03 17:22:10.9350 [RESULT#-1872162716 ap_21fe11ad_B3_P0_00194_20120501_26398.wu_0] Valid; granted 19.657693 credit [HOST#1504137] 2012-05-03 17:22:10.9384 [debug] [HAV#12000003] consecutive valid now 75 2012-05-03 17:22:10.9412 [RESULT#-1872162715 ap_21fe11ad_B3_P0_00194_20120501_26398.wu_1] Valid; granted 19.657693 credit [HOST#5634961] The print outs are truncated to 32 bit signed numbers so result -1872162716 is 2422804580 and -1872162715 is 2422804581. For some your wingman's anonymous platform numbers credit claims are way off, even though yours are pretty close to spot on. But anonymous vs anonymous always grants the low number. (Another thing I'd like to fix.) @SETIEric@qoto.org (Mastodon) |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.