clustering


log in

Advanced search

Questions and Answers : Wish list : clustering

Author Message
Matthew Springett
Send message
Joined: 16 Jul 99
Posts: 1
Credit: 136,130
RAC: 330
United States
Message 1300609 - Posted: 31 Oct 2012, 0:33:12 UTC

It would be an advantage to be able to link several computers together to work on the same work unit at the same time. Or have one multi-core computer use multiple cores on a single work unit.
If this feature is already available, it is not clear as to how to implement it.
____________

John McLeod VII
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 15 Jul 99
Posts: 24544
Credit: 522,500
RAC: 102
United States
Message 1300627 - Posted: 31 Oct 2012, 2:31:10 UTC - in response to Message 1300609.

It would be an advantage to be able to link several computers together to work on the same work unit at the same time. Or have one multi-core computer use multiple cores on a single work unit.
If this feature is already available, it is not clear as to how to implement it.

Clustering, not really. However, if the application can use multi threads, and it has been set up for that, then BOINC is perfectly happy to use multiple threads on a single task. However, SETI is only single threaded.
____________


BOINC WIKI

Profile eteela
Send message
Joined: 27 Aug 12
Posts: 16
Credit: 2,065,445
RAC: 1,573
United States
Message 1303923 - Posted: 9 Nov 2012, 10:06:09 UTC - in response to Message 1300609.
Last modified: 9 Nov 2012, 10:07:40 UTC

It might seem like an advantage, but the more you split the problem the more overhead there is. You could see a WU process faster by splitting it further, but you would be working on fewer units at a time. You'd still end up with the same RAC, or a little less because of the extra overhead.

I can see a little use for this where you want to get a WU done fast then stop processing for a while. But this project is really set up for a MUCH longer term and that situation does not occur frequently enough to concider due to the overhead involved. Some GPU's already process these WU's too quickly.

In regards to seti@home the problem is already being run on multiple cores in a cluster. That's BOINC. The WU is already a very small portion of the whole problem.

for an example...

time in seconds
total time / cpu time / credit
2,638. / 1,612. / 732.59 AstroPulse v6 (NVIDIA GPU) GeForce GTX 690
520,008. / 452,082. / 732.59 AstroPulse v6 v6.02 (Core2 Duo CPU @ 2.40GHz) macbook 2008
http://setiathome.berkeley.edu/workunit.php?wuid=1053500772
Isn't 44 miniutes fast enough for a AstroPulse WorkUnit? I'm sure the guy with the macbook thinks so after 5 days.

Questions and Answers : Wish list : clustering

Copyright © 2014 University of California