Scheduler problem??


log in

Advanced search

Message boards : Number crunching : Scheduler problem??

Author Message
rob smithProject donor
Volunteer tester
Send message
Joined: 7 Mar 03
Posts: 8535
Credit: 59,470,296
RAC: 86,463
United Kingdom
Message 1366115 - Posted: 9 May 2013, 19:28:11 UTC

Recently I've noticed that one of my crunchers will be happily "bouncing off the limits". It will then report a few tasks, and thus request a few new tasks. The server responds with "Scheduler request completed: got 0 new tasks", followed quickly by "No tasks sent", messages about type of task not available, then finally "This computer has reached a limit on tasks in progress". The last statement is incorrect, as a quick look see shows that the computer no longer has 200 tasks "in progress", but somewhat less. This will repeat for a couple of cycles before it gets enough tasks to return it to 200 in progress. This appears to most commonly happen when there are about 10 tasks reported in one cycle.
____________
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Horacio
Send message
Joined: 14 Jan 00
Posts: 536
Credit: 75,118,405
RAC: 38,495
Argentina
Message 1366163 - Posted: 9 May 2013, 22:59:56 UTC - in response to Message 1366115.

It may get the limit reached message even if there are less than 200 because the limits are not really 200, its 100 for GPU and 100 for CPU, and when one of them is reached you get that message if the client was asking for both kind of work or for the one that is in the limit...

Most common situation is when the CPU has enough work for the cache size choosen (and this amount could be less than 100 if your cache is not set to a very hughe size or there are not so many CPU cores in that host) and then the only real limit in effect is the one for the GPU.
____________

Lionel
Send message
Joined: 25 Mar 00
Posts: 576
Credit: 236,073,634
RAC: 230,919
Australia
Message 1366168 - Posted: 10 May 2013, 0:04:23 UTC - in response to Message 1366115.

Rob, go to the Projects tab and highlight Seti@home. When your communication deferred time gets downs to less than 3 seconds, click on the Update button on the Projects tab. If there is work, you should see both GPU and CPU WUs coming through.
____________

Message boards : Number crunching : Scheduler problem??

Copyright © 2014 University of California