Only one core used
14 Jan 2009 16:24:24 UTC
Topic 194139
(moderation:
Hi,
I have a system with 2 cores and 2 projects (Einstein and LHC). Up to about dec I used to have 2 Einstein WU's crunched at any point in time since LHC only occassionaly offers WU's.
Recently, however, only 1 Einstein WU is being crunched on one core and the other core remains "idle" because LHC does not offer WU's. In my messages I can see that only LHC WU's are asked for. Einstein WU's are only asked for when the running WU is near 100%.
What do I need to change to get 2 Einstein WU's running?
regards
Only one core used
)
It sounds like you need to go to your lhc@home preferences and turn your resource share for lhc@home down from 100 to something more like 25 or less. Right now Einstein probably owes lhc a good deal of debt due to the lack of lhc tasks, adjusting your resource share should help with that.
~It only takes one bottle cap moving at 23,000 mph to ruin your whole day~
Hi Mary, Good suggestion!
)
Hi Mary,
Good suggestion! I changed it and immediately Einstein work was asked. Now I have 2 Einstein WU's running.
Before, I have tried already a lot of things including setting LHC to "No new tasks". Nothing worked.
thanks
RE: ...Before, I have tried
)
Perhaps suspending LHC would have worked. I'm not sure, but I think there's a difference between NNT and suspend regarding debt calculation.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
Mary's suggestion will help
)
Mary's suggestion will help for a while but eventually Einstein's debt will probably grow again to the point where the problem returns. It depends on how much work you get from LHC. If the problem returns you should reset the debts to zero by detaching and reattaching LHC or Einstein.
BOINC FAQ Service
Official BOINC wiki
Installing BOINC on Linux
RE: ...If the problem
)
...or by using a tool. (I use a downloaded version)
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
RE: Mary's suggestion will
)
Setting a project to 'No new tasks' prevents you from owing any more debt to that project than you already do (I have to do this in the case of STZAKI), but it doesn't necessarily free up the resource share you have set for the 'NNT' project. That's where changing the resource share comes in.
Suspending a project frees up its resource share, but I haven't checked to see what happens in regards to debt with suspended projects in the long run. Regardless, it should take quite a while to rack up enough debt to cause this problem again as long as the resource allocation for LHC is set fairly low since LHC tasks pop up ~every two weeks or so (although the last batch was quite large, so they may very well be ramping back up). And setting LHC to 'NNT' when the server status page shows no available WU's can prevent this debt accumulation almost entirely.
~It only takes one bottle cap moving at 23,000 mph to ruin your whole day~
RE: RE: Mary's suggestion
)
Sorry, my post made it sound like there is little or no need for maris to reduce LHC's resource share. I am in total agreement with you on that point, s/he definitely needed to reduce LHC's share.
I wrote a Python script that obtains LTDs via RPC calls periodicly, saves them in a table, and calls gnuplot to plot the data on a graph. Such script might be useful if you want to study it over the long run.
Agreed. It will take quite a while and that's why I said "eventually".
But, other than men of leisure like me, who can watch the server status page all the time? Entire batches can disappear in a matter of a few hours. If I were forced to a manual solution, I would rather just reset the debts once in a while though the solution one finds easiest depends on many factors. Keeners, on the other hand, have a script that watches server status for them and manipulates suspend/resume and NNT accordingly or else resets LTDs when they reach a certain level.
BOINC FAQ Service
Official BOINC wiki
Installing BOINC on Linux
Hi, Thanks for the
)
Hi,
Thanks for the details. Resetting the debt also looks like a solution.
At this point I really wonder what is the use/benefit of the Debt parameter.
If I run 2 projects with equal share, then whenever both projects offer WU's, I will have running WU's from both projects. Right?
If one project does not have WU's why should I prevent the other project running additional WU's?
Can someone explain the design and goal of the Debt parameter?
Anyway, it looks like the project share parameters are tied to the total sum of the time that projects have run in the past, rather than to only the present time. The latter makes more sense in my opinion.
regards
RE: Hi, Thanks for the
)
It is in there because projects like Climate Prediction have units that can take 2 years to complete and they use alot of the cpu time. That is alot of time crunching just for them and not for your other favorite projects. After you finish one of those units, not all of their units are so large, you may not crunch for them again for a month or so.
RE: Hi, Thanks for the
)
It's just a way of counting the time each of your projects has been crunched, for the purpose of giving each of your projects the share you specify.
Usually but not always. It depends on your cache size, time between connections, deadlines and other factors. When you think about it, you don't really have to have a task from each project running (or at least in the cache if not running) at all times if you have a debt accounting system that makes sure no project gets more than its share over the longterm. Remember that another important goal is that you don't want to download tasks that will not meet the deadline.
You should not prevent the other project from running additional tasks. Neither should the scheduler but unfortunately there is a flaw in way the scheduler is implemented. Due to that flaw, if you are attached to a project that doesn't give you enough work to meet the resource share set for that project then eventually the scheduler will stop requesting work for other projects. The developers are working to fix that flaw but it's a very difficult problem. Until it's perfected, we need to use the workarounds described earlier (setting a realistic resource share and/or reseting the debts occassionally) if we attach projects that don't supply steady work.
The Client CPU/GPU scheduling and Work fetch and GPUs links in the section titled Design docs for current development in the Software Development document might answer that question the best.
BOINC FAQ Service
Official BOINC wiki
Installing BOINC on Linux