Message boards :
Questions and problems :
Underutilisation after upgrading to 6.10.58
Message board moderation
Previous · 1 · 2
Author | Message |
---|---|
Send message Joined: 25 Oct 10 Posts: 13 |
That was my first thought to. After looking had enough work, but not over SINGLE projects ie. project 1: 0 jobs project 2: 3 jobs project 3: 10 jobs project 4: 8 jobs project 5: 0 jobs project 6: 2 jobs project 7: 4 jobs project 8: 9 jobs project 9: 6 jobs project 10: 2 jobs Now since i have told the program to change between projects in XX minutes and it does, if there is not enough work in the current (active rotated) project it should do work anyway for the next project as well as the (active rotated) project. Now if this is the case there is a fundamental flaw in the program methodology hence Underutilisation after upgrading to 6.10.58 Regards Robert |
Send message Joined: 29 Aug 05 Posts: 15483 |
I did see that your CPDN model was overdue (or thought to be overdue in due time), thus running in high priority. At such time that core will not be used by any other project. And I wonder here if that has any impact on what work will be downloaded from the other project or started on the other CPU cores. Anyway, I was chatting with one of the developers tonight, I've warned him about this thread already. I'll pass my suspicions about this by him tomorrow, after I had some sleep. :) |
Send message Joined: 25 Oct 10 Posts: 13 |
I think the 2 images already sent validates our observations. Again thanks for all your help |
Send message Joined: 29 Aug 05 Posts: 15483 |
OK, I am pointing out some things to the developers through email. Do know that they live in the US, one on the east coast, the other on the west coast, so varying times they're available. I'm also wondering about the multi-threaded AQUA application, in how far it can play havoc when being scheduled. It'll want to use all CPU cores... which it won't be able to do for a while yet, since that CPDN task is in high priority. Can you check if either of those AQUA's have run in the mean time, and on how many cylinde.. err cores? :-) |
Send message Joined: 25 Oct 10 Posts: 13 |
Who said life was meant to be easy. In the old version even with the high priority, there was never a problem, always shared resources. Looking at the % completed it seems to be working. all processors at 100% http://www.rditsolutions.com.au/seti/AQUA@home.txt |
Send message Joined: 29 Aug 05 Posts: 15483 |
OK, I have a request. The moment you see your BOINC run on less than all of its cores again, 4, 2, whatever, please email me immediately. I'll PM you my email address. Add into that email your client_state.xml file, your global_prefs.xml file and if you have one, your cc_config.xml file. Please zip or otherwise compress them before sending them to me. I will then run your work through the BOINC simulator to see if I can reproduce your problem. With thanks. |
Send message Joined: 3 Apr 06 Posts: 547 |
Are you possibly talking about the problem which occasionally happens on systems, where a multi-CPU AQUA task is waiting to be processed and sometimes less than nCPUs are utilized by other tasks? It has been mentioned recently on the Alpha mailing list. IIRC the devs meant it happens due to AQUA calling incorrect initialization routine. Peter |
Send message Joined: 3 May 09 Posts: 5 |
I notice that : when a wu is running in Higth Priority, at end of one other wu, the core this ending wu use is not affected to a new wu. At end stay running only the Hight Priority wu and a gpu milkyway are running. I suspend the wu in Higth Prioryty, an aqua wu start and use my 8 cores i7, after the timeslot for aqua, 8 wus restarts. |
Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License,
Version 1.2 or any later version published by the Free Software Foundation.