Project to run while running out cache
18 Jan 2009 5:07:34 UTC
Topic 194145
(moderation:
My s5r4 caches are getting close to fully depleted, and I need something to run on idle cores while the last r4 WUs are completed.
THe last time I did this, I used primegrid, but the boinc sceduler's not cooperating today. My resource share is 100:1. It's apparently decided to go in EDF mode and since hte prime grid WUs have deadlines of only a few days is running them exclusively.
Can anyone recommend a project with reasonably short WUs that will behave better with the scheduler?
Project to run while running out cache
)
Dan,
Part of the problem, in my opinion, is that the lopsided ratio is partly to blame. Though many of the experts claim that BOINC Manager will work well in this situation, my experience is that it does not ...
What I mean is that a lot of the time it will behave and will do pretty much what you want ... but, every now and then it will get into a snit and stop doing things sensibly. I think this is partly because the main testing mode is based on fairly equal resource shares and ignoring short term hiccups.
For example I had ABC set to run 40% ... and I know that I had not run but one ABC task for quite a few hours so it should have been requesting more tasks ... I reset all LTD to zero, which should have created a perfect balance so that it would have gotten more work ... nope... ask for 0 seconds ... it was not like the task was in deadline jeopardy ... and I had 7 more cores running ...
Anyway ...
I would have suggested Malaria, but they are now in intermittent mode as are so many other projects ... Rosetta is also having on and off problems and serious overruns on tasks ...
About an hour: Leiden, POEM, Enigma, Docking
Two to three hours: AI, WCG (selected sub-projects, not CEP), Genetic Life
RE: My s5r4 caches are
)
Yes, S5R5 does the job nicely :-).
I was in the process of running down the caches on 150 machines all of which were running under the AP mechanism so normally you wait for the cache to drain before deleting app_info.xml (otherwise you lose all remaining tasks). I have developed a way to do the transition before the cache is exhausted so that new R5 tasks become available and no core needs to run dry. I've converted over 100 machines so far and none had exhausted their R4 work and none lost any of the tasks as a result of the surgery. All these machines now have dual R4/R5 tasks in their cache and all have been taken off NNT.
The method is quite straight forward but is a little tedious as it involves careful editing of the state file. If you would like detailed instructions just ask. You do need one machine that is running the official R4 app (ie 6.02 for Linux or 6.10 for Windows) so you can capture the file signatures for the various apps that make up the R4 app package. These file signatures are what prevent things going awry when you delete app_info.xml
Cheers,
Gary.