rpi_logo
scheduler update
scheduler update
log in

Advanced search

Message boards : News : scheduler update

Author Message
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0

Message 43479 - Posted: 4 Nov 2010, 21:07:36 UTC

I updated the scheduler, so hopefully it should be correctly sending out SSE2 applications. Let me know if it's working here.
--Travis
____________

kordoch
Send message
Joined: 15 Feb 10
Posts: 1
Credit: 159,982
RAC: 0

Message 43505 - Posted: 5 Nov 2010, 13:34:13 UTC

SSE2 is working for me. I got two work units a couple hours ago.

-kordoch

Profile John Black
Send message
Joined: 3 May 10
Posts: 74
Credit: 1,532,760
RAC: 0

Message 43511 - Posted: 5 Nov 2010, 15:52:31 UTC - in response to Message 43479.

Hi Travis
everything seems to be back to normal if thats the right word SSE2 apps running well. Once BOINC gets the runtime right it will really be back to normal. Is there any downtime planned this week?

mfbabb2
Send message
Joined: 29 Sep 09
Posts: 18
Credit: 46,059
RAC: 0

Message 43523 - Posted: 5 Nov 2010, 20:02:12 UTC

Does this mean it will properly send NON-SSE2 WU's to the old AMD machines which do NOT support SSE2?
____________

Matt Arsenault
Volunteer moderator
Project developer
Project tester
Project scientist
Send message
Joined: 8 May 10
Posts: 576
Credit: 15,979,383
RAC: 0

Message 43534 - Posted: 6 Nov 2010, 1:32:05 UTC - in response to Message 43523.

Does this mean it will properly send NON-SSE2 WU's to the old AMD machines which do NOT support SSE2?


That should have been fixed a long time ago. The patch to the scheduler to actually send out SSE2 ones was apparently not applied before.

Marc Lewis
Send message
Joined: 1 Jan 09
Posts: 1
Credit: 1,251,478
RAC: 0

Message 43549 - Posted: 6 Nov 2010, 14:48:03 UTC - in response to Message 43479.

I updated the scheduler, so hopefully it should be correctly sending out SSE2 applications. Let me know if it's working here.<br>--Travis


It seems to be okay. Here is the result from my boinc log:
=========CUT==========
11/6/2010 9:40:26 AM Milkyway@home Scheduler request completed: got 1 new tasks
11/6/2010 9:40:29 AM Milkyway@home Started download of milkyway_0.45_windows_intelx86__sse2.exe
11/6/2010 9:40:29 AM Milkyway@home Started download of msvcp100.dll
11/6/2010 9:40:31 AM Milkyway@home Finished download of milkyway_0.45_windows_intelx86__sse2.exe
11/6/2010 9:40:31 AM Milkyway@home Started download of msvcr100.dll
11/6/2010 9:40:33 AM Milkyway@home Finished download of msvcp100.dll
11/6/2010 9:40:33 AM Milkyway@home Started download of p-82-3s-edge.txt
11/6/2010 9:40:34 AM Milkyway@home Finished download of p-82-3s-edge.txt
11/6/2010 9:40:34 AM Milkyway@home Started download of stars-td82-2stream_40.txt
11/6/2010 9:40:36 AM Milkyway@home Finished download of msvcr100.dll
11/6/2010 9:40:36 AM Milkyway@home Started download of de_separation_82_3s_40_2_1107457_1289047887_search_parameters
11/6/2010 9:40:37 AM Milkyway@home Finished download of de_separation_82_3s_40_2_1107457_1289047887_search_parameters
11/6/2010 9:40:47 AM Milkyway@home Finished download of stars-td82-2stream_40.txt
11/6/2010 9:40:47 AM Milkyway@home Starting de_separation_82_3s_40_2_1107457_1289047887_1
11/6/2010 9:40:47 AM Milkyway@home Starting task de_separation_82_3s_40_2_1107457_1289047887_1 using milkyway version 45
11/6/2010 9:41:28 AM Milkyway@home Sending scheduler request: To fetch work.
11/6/2010 9:41:28 AM Milkyway@home Requesting new tasks
11/6/2010 9:41:33 AM Milkyway@home Scheduler request completed: got 1 new tasks
11/6/2010 9:41:35 AM Milkyway@home Started download of milkyway_nbody_0.21_windows_intelx86__sse2.exe
11/6/2010 9:41:35 AM Milkyway@home Started download of orphan_model4_32768_bodies.js
11/6/2010 9:41:36 AM Milkyway@home Finished download of orphan_model4_32768_bodies.js
11/6/2010 9:41:36 AM Milkyway@home Started download of histogram_corrected_2
11/6/2010 9:41:38 AM Milkyway@home Finished download of histogram_corrected_2
11/6/2010 9:41:46 AM Milkyway@home Finished download of milkyway_nbody_0.21_windows_intelx86__sse2.exe
11/6/2010 9:41:46 AM Milkyway@home Starting de_nbody_model4_3_41374_1289052102_0
11/6/2010 9:41:46 AM Milkyway@home Starting task de_nbody_model4_3_41374_1289052102_0 using milkyway_nbody version 21
=========CUT==========



____________

dpost
Send message
Joined: 13 Oct 10
Posts: 1
Credit: 4,570,438
RAC: 1,218

Message 43553 - Posted: 6 Nov 2010, 18:23:31 UTC

I have not gotten any tasks to run for several days. Request for tasks gives two, but they do not run. CPU is idle. I run windows 2003 server with 2 dual core xenon processors. Dick

Profile Danger
Send message
Joined: 20 Feb 10
Posts: 3
Credit: 5,030,476
RAC: 0

Message 43573 - Posted: 7 Nov 2010, 14:52:27 UTC

Running 4 SSE2 apps currently, all looks good.


-Zuke
"I'm not slacking, my code is compiling."

Bofferbrauer
Send message
Joined: 19 Jul 10
Posts: 3
Credit: 586,790
RAC: 3,048

Message 43578 - Posted: 7 Nov 2010, 19:49:22 UTC

Working fine on my computer, the only issue is the expected runtime being around three times higher as the resulting runtime(told me taking 33h but finished in 10 and a half hour).
____________

David C!
Avatar
Send message
Joined: 30 Oct 09
Posts: 4
Credit: 76,212
RAC: 0

Message 43579 - Posted: 7 Nov 2010, 20:19:17 UTC
Last modified: 7 Nov 2010, 20:21:03 UTC

IT'S WORKING AGAIN! I had to Reset Project to get it going. It's good to see that MW is working again; however, 27 hour work units with less than 8 days to completion forces Boinc into high priority processing dedicated to MW. Can't the WU's be broken down into smaller units of time?

Matt Arsenault
Volunteer moderator
Project developer
Project tester
Project scientist
Send message
Joined: 8 May 10
Posts: 576
Credit: 15,979,383
RAC: 0

Message 43580 - Posted: 7 Nov 2010, 22:21:34 UTC - in response to Message 43579.

IT'S WORKING AGAIN! I had to Reset Project to get it going. It's good to see that MW is working again; however, 27 hour work units with less than 8 days to completion forces Boinc into high priority processing dedicated to MW. Can't the WU's be broken down into smaller units of time?

Are these N-body workunits? The N-body is hard to predict the actual runtime. Over the range of parameters being fit, there is a huge potential for variation. If the estimate isn't high enough, BOINC kills them off prematurely. I've sort of been working on a better, and we've been talking about maybe doing partial N-body simulations in a single workunits (which will be necessary if more bodies are required) although that would make everything more complicated.

Bofferbrauer
Send message
Joined: 19 Jul 10
Posts: 3
Credit: 586,790
RAC: 3,048

Message 43649 - Posted: 9 Nov 2010, 23:44:18 UTC - in response to Message 43580.

IT'S WORKING AGAIN! I had to Reset Project to get it going. It's good to see that MW is working again; however, 27 hour work units with less than 8 days to completion forces Boinc into high priority processing dedicated to MW. Can't the WU's be broken down into smaller units of time?

Are these N-body workunits? The N-body is hard to predict the actual runtime. Over the range of parameters being fit, there is a huge potential for variation. If the estimate isn't high enough, BOINC kills them off prematurely. I've sort of been working on a better, and we've been talking about maybe doing partial N-body simulations in a single workunits (which will be necessary if more bodies are required) although that would make everything more complicated.


My last N-body announced 44h of work... finally it was done in less than 2h. I understand that that the actual runtime may be hard to prdict, but such a difference look laughable
____________

Matt Arsenault
Volunteer moderator
Project developer
Project tester
Project scientist
Send message
Joined: 8 May 10
Posts: 576
Credit: 15,979,383
RAC: 0

Message 43650 - Posted: 10 Nov 2010, 0:06:32 UTC - in response to Message 43649.

My last N-body announced 44h of work... finally it was done in less than 2h. I understand that that the actual runtime may be hard to prdict, but such a difference look laughable

The workunits need an upper bound for the number of FLOPs needed. Right now the same number I think is used for all in a set of workunits. Over the parameters being fit, the individual workunits vary by a large factor, and there is an element of randomness as well. The estimates need to be refined for individual workunits.

Optikus
Send message
Joined: 15 Aug 10
Posts: 6
Credit: 41,856
RAC: 0

Message 43751 - Posted: 11 Nov 2010, 20:06:31 UTC


Post to thread

Message boards : News : scheduler update


Main page · Your account · Message boards


Copyright © 2018 AstroInformatics Group