Message boards :
ATLAS application :
highly variable credit points
Message board moderation
Author | Message |
---|---|
Send message Joined: 18 Dec 15 Posts: 1808 Credit: 118,207,403 RAC: 29,480 |
This morning I noticed a - at least what I think - major imbalance in credits allocation between 2 PCs on which I am running ATLAS: 1-core task, system with old 2.83GHz processor: runtine: 18,828.02 - CPU time: 18,088.45 - credits: 342.58 2-core task, system with modern 3.4GHz processor: runtime: 16,046.83 - CPU time: 30,308.48 - credits: 298.68 so my question is: what sense does it make to run a 2-core task, if a 1-core task, within same time-span, yields same credit? |
Send message Joined: 18 Dec 16 Posts: 123 Credit: 37,495,365 RAC: 0 |
what sense does it make to run a 2-core task, if a 1-core task, within same time-span, yields same credit? Inconsistency in credit allocation has been noticed before. The rational behind the algorithm used for credit allocation is outside the control of the LHC@Home team from my understanding. However, it seems that you have just recently started using your 2.83GHz processor for crunching ATLAS tasks. Most likely, your credit will gradually decrease over time. Wait a couple of days. We are the product of random evolution. |
Send message Joined: 28 Sep 04 Posts: 728 Credit: 48,797,412 RAC: 25,062 |
Also single core tasks are getting odd credits at the moment. Here is a list of my six last ATlas tasks: 140816763 68071035 15 May 2017, 23:02:58 UTC 17 May 2017, 16:30:35 UTC Completed and validated 42,166.30 41,455.55 121.28 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 140213407 67772122 14 May 2017, 19:29:48 UTC 17 May 2017, 7:10:25 UTC Completed and validated 40,146.21 39,521.96 179.36 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 140191101 67761034 14 May 2017, 18:07:44 UTC 17 May 2017, 3:29:24 UTC Completed and validated 42,940.68 42,354.01 226.74 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 140166619 67748830 14 May 2017, 17:14:12 UTC 16 May 2017, 18:47:40 UTC Completed and validated 42,581.75 41,796.40 302.93 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 140166610 67748824 14 May 2017, 17:00:52 UTC 16 May 2017, 12:46:44 UTC Completed and validated 46,429.92 44,928.74 355.77 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 140120393 67726955 14 May 2017, 15:06:11 UTC 16 May 2017, 4:37:21 UTC Completed and validated 42,012.12 40,773.95 330.69 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 All are around 40000 seconds but credit has dropped from 300+ to 120+ |
Send message Joined: 28 Sep 04 Posts: 728 Credit: 48,797,412 RAC: 25,062 |
Next one finished, about 40000 seconds and credit = 107. |
Send message Joined: 18 Dec 15 Posts: 1808 Credit: 118,207,403 RAC: 29,480 |
Next one finished, about 40000 seconds and credit = 107. by tomorrow, credit will be ZERO :-) |
Send message Joined: 27 Aug 15 Posts: 27 Credit: 12,957,000 RAC: 26,012 |
1-core credit for 8-core tasks??? Where are the credits? 141990828 68608803 10474143 17 May 2017, 14:48:01 UTC 18 May 2017, 5:25:06 UTC Fertig und Bestätigt 3,925.97 15,403.50 85.47 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 141576619 68448201 10474143 17 May 2017, 0:01:43 UTC 18 May 2017, 3:36:42 UTC Fertig und Bestätigt 3,952.93 15,285.05 84.94 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 141576622 68448203 10474143 17 May 2017, 0:01:43 UTC 18 May 2017, 4:44:25 UTC Fertig und Bestätigt 4,074.87 15,621.75 87.67 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 141534152 68426899 10474143 16 May 2017, 22:57:24 UTC 18 May 2017, 2:30:05 UTC Fertig und Bestätigt 4,704.83 15,622.61 101.99 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 141477626 68398503 10474143 16 May 2017, 21:24:19 UTC 18 May 2017, 0:08:53 UTC Fertig und Bestätigt 3,762.59 13,871.59 85.88 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 141435144 68377294 10474143 16 May 2017, 19:33:41 UTC 17 May 2017, 23:02:46 UTC Fertig und Bestätigt 3,864.57 15,046.31 90.22 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 141435041 68377221 10474143 16 May 2017, 19:33:41 UTC 18 May 2017, 0:08:53 UTC Fertig und Bestätigt 3,882.46 15,061.89 87.86 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 141375573 68347703 10474143 16 May 2017, 17:58:00 UTC 17 May 2017, 21:22:43 UTC Fertig und Bestätigt 3,654.06 16,082.56 87.07 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 141312411 68316266 10474143 16 May 2017, 17:07:40 UTC 17 May 2017, 18:59:58 UTC Fertig und Bestätigt 4,313.91 16,040.38 107.03 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 141312226 68316170 10474143 16 May 2017, 17:07:40 UTC 17 May 2017, 20:44:33 UTC Fertig und Bestätigt 3,645.95 15,795.19 87.62 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 141208590 68264626 10474143 16 May 2017, 14:08:22 UTC 17 May 2017, 17:53:51 UTC Fertig und Bestätigt 3,913.84 13,802.05 98.77 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 141206814 68263800 10474143 16 May 2017, 12:58:26 UTC 17 May 2017, 16:51:43 UTC Fertig und Bestätigt 6,517.53 28,792.34 167.76 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 141157612 68183368 10474143 16 May 2017, 11:59:31 UTC 17 May 2017, 14:48:01 UTC Fertig und Bestätigt 4,575.44 14,482.95 125.04 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 139617653 66811812 10474143 16 May 2017, 11:00:36 UTC 16 May 2017, 16:27:42 UTC Fertig und Bestätigt 3,411.46 13,434.58 235.85 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 141029271 68176360 10474143 16 May 2017, 8:51:41 UTC 17 May 2017, 13:33:39 UTC Fertig und Bestätigt 4,164.97 12,952.20 117.74 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 141014853 68169183 10474143 16 May 2017, 8:36:23 UTC 17 May 2017, 12:11:35 UTC Fertig und Bestätigt 4,515.94 14,253.73 138.04 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 141016168 68073723 10474143 16 May 2017, 8:36:23 UTC 17 May 2017, 12:53:16 UTC Fertig und Bestätigt 4,414.51 14,103.16 128.98 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 140939601 68131761 10474143 16 May 2017, 5:29:52 UTC 17 May 2017, 9:52:52 UTC Fertig und Bestätigt 4,456.13 14,185.97 156.57 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 139477749 67400736 10474143 16 May 2017, 4:27:28 UTC 16 May 2017, 8:51:40 UTC Fertig und Bestätigt 4,113.52 16,324.28 300.30 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 140903083 68113598 10474143 16 May 2017, 3:27:29 UTC 17 May 2017, 8:40:40 UTC Fertig und Bestätigt 4,254.88 15,176.91 159.05 ATLAS Simulation v1.01 (vbox64_mt_mcore_atlas) windows_x86_64 |
Send message Joined: 2 Sep 04 Posts: 455 Credit: 201,059,850 RAC: 37,603 |
|
Send message Joined: 28 Sep 04 Posts: 728 Credit: 48,797,412 RAC: 25,062 |
I wonder if the long estimated runtimes are effecting the credits. The tasks usually finish around 11 hours (1 core) but now the estimates are 62 hours and they drop very slowly, after 7 hours (54 events finished) still expected to last 58.5 hours. I'll keep my eye on it to see how it goes. |
Send message Joined: 15 Jun 08 Posts: 2528 Credit: 253,722,187 RAC: 73,124 |
1-core credit for 8-core tasks??? Where are the credits? 85.47 credits for 3,925.97 s runtime seems to be very low for that kind of host. On the other hand 3,925.97 s runtime and only 15,403.50 CPU time also seems to be very unefficient (49 %). You may test a different setup, e.g. 2x 4-core, 2x 3-core or 3x 2-core. Let them run a couple of WUs (>10 each) to get the credit calculation stable. |
Send message Joined: 15 Jun 08 Posts: 2528 Credit: 253,722,187 RAC: 73,124 |
There may be a wrong (much too high) setting for <rsc_fpops_est>. Unfortunately I can not check this for ATLAS as the server preferred to send me other subprojects for a couple of days. The <rsc_fpops_est> affects the client's runtime calculation (therefore also other projects!) as well as - with a delay - the credit calculation and has to be set by the project as close as possible to the real runtime (see BOINC wiki). |
Send message Joined: 13 May 14 Posts: 387 Credit: 15,314,184 RAC: 0 |
You are right, we were recently putting a fpops_est way higher than it should have been. We normally set it to 4 hours * 3GHz, assuming the average 3GHz processor can finish a single core task in 4 hours. But a small problem caused this to go wrong in the last few days. I have changed all the bad values in the BOINC database so estimates should go back to normal and so credit should (slowly) adjust back to the normal values. |
Send message Joined: 28 Sep 04 Posts: 728 Credit: 48,797,412 RAC: 25,062 |
Good to know that a solution for this was found. But the value of 4 hours * 3 GHZ seems very optimistic. I run a i7-3770 CPU at ~3.7 GHz (Intel Turbo Boost) and a single core task takes usually 10-12 hours. |
Send message Joined: 2 May 07 Posts: 2240 Credit: 173,894,884 RAC: 3,757 |
Tasks with 2 CPU's (5GByte Memory) and SSD-disk Computer between 3.2 and 3.7 GHz. ATLAS (43.200 GFlops) need about 4 hours. |
Send message Joined: 18 Dec 15 Posts: 1808 Credit: 118,207,403 RAC: 29,480 |
But the value of 4 hours * 3 GHZ seems very optimistic. I run a i7-3770 CPU at ~3.7 GHz (Intel Turbo Boost) and a single core task takes usually 10-12 hours. I fully agree, based on my own, long-time experience |
Send message Joined: 28 Sep 04 Posts: 728 Credit: 48,797,412 RAC: 25,062 |
Tasks with 2 CPU's (5GByte Memory) and SSD-disk That is very fast, based on Boinc's CPU benchmarks my i7-3770 should be a little faster than your AMD FX-8370E. Have you figured out what is the major factor for that speed? 2 CPU cores, 5 GByte memory, SSD disk or what? Your network speed seems also a lot faster than mine (10/100 Mbits/s). |
Send message Joined: 2 Sep 04 Posts: 455 Credit: 201,059,850 RAC: 37,603 |
You are right, we were recently putting a fpops_est way higher than it should have been. We normally set it to 4 hours * 3GHz, assuming the average 3GHz processor can finish a single core task in 4 hours. But a small problem caused this to go wrong in the last few days. I have changed all the bad values in the BOINC database so estimates should go back to normal and so credit should (slowly) adjust back to the normal values. David, why do you stay with estimates ? You have a database with years of experience how long it takes to crunch a defined amount of work. You could analyse this database and figure out values based on real well known facts Supporting BOINC, a great concept ! |
Send message Joined: 2 May 07 Posts: 2240 Credit: 173,894,884 RAC: 3,757 |
VDSL (80/40 MBits). Up and download of a task take about 2 Minutes. The SSD? Don't know what's the reason. Have also a AMD A10 with SSD. Needs the same time. No overclocking. |
Send message Joined: 27 Sep 08 Posts: 844 Credit: 690,642,842 RAC: 114,313 |
There is also the benchmark application from Laurence that would be good source of data or could be modified for Atlas application |
Send message Joined: 13 May 14 Posts: 387 Credit: 15,314,184 RAC: 0 |
You are right, we were recently putting a fpops_est way higher than it should have been. We normally set it to 4 hours * 3GHz, assuming the average 3GHz processor can finish a single core task in 4 hours. But a small problem caused this to go wrong in the last few days. I have changed all the bad values in the BOINC database so estimates should go back to normal and so credit should (slowly) adjust back to the normal values. That is true but every group of tasks is different and sometimes there are different numbers of events in the WU. We would have to make a prediction for each WU based on the history of previous WU for that particular group of tasks. It would be doable but would take some work. I think that the 4h * 3GHz estimation was based on the old ATLAS tasks with 50 events per WU so maybe we can increase it. My own host gets quite close - 4500s for 4-core WU, but it has the unfair advantage of being inside the CERN network :) |
Send message Joined: 18 Dec 15 Posts: 1808 Credit: 118,207,403 RAC: 29,480 |
another case of major difference in credits: task 153600868 - runtime 12.679 secs - CPU time 22.983 secs - points: 250,92 task 153601314 - runtime 12.920 secs - CPU time 24.032 secs - points: 147,67 both 2-core tasks were from August 13. I have seen this also with tasks from the days before: some yielded higher points, some lower points. any explanation for this inconsistency? |
©2024 CERN