Message boards :
Number crunching :
Preempting result 008b_300025284_0 (quit)
Message board moderation
Author | Message |
---|---|
Send message Joined: 25 Aug 04 Posts: 1 Credit: 299,454 RAC: 0 |
SETI has gone live on BOINC 4.05 as soon as it download the new client process and WU\'s climateprediction displayed the following. climateprediction.net - 2004-08-29 20:08:15 - Preempting result 008b_300025284_0 (quit) Any suggestion?? Thanks in advance. Kevin.... |
Send message Joined: 7 Aug 04 Posts: 27 Credit: 766,972 RAC: 0 |
It's the new time sharing features of boinc 4.x clients. What is currently happening is every 1 hour, boinc pauses the crunching for one project (preempt) and crunches a work unit from another project. After 1 hour again, it'll preempt that work unit and will to to the third project (or back to the first project if you're only attached to 2). This way, those seti work units don't sit in your queue for 2 months waiting for the cpdn work unit to finish. --- <IMG SRC="http://www.funkymonkey.org/boinc/sig.php?id=925&proj=pah"> <IMG SRC="http://www.funkymonkey.org/boinc/sig.php?id=169421&proj=sah"> <IMG SRC="http://www.funkymonkey.org/boinc/sig.php?id=360&proj=cpdn"> |
Send message Joined: 5 Aug 04 Posts: 907 Credit: 299,864 RAC: 0 |
well I think it's not quite every hour, it depends on your user preferences for each project, i.e. if you set CPDN up as a "resource share" of 100 and SETI as 100 they will share equally. I hope it's no less than an hour, because if people have "leave suspended apps in memory" to false that will hit a lot of lost checkpoint work (a CPDN checkpoint is every 144 timesteps, which can be 10-15 minutes or longer between checkpoints depending on computer speed). |
Send message Joined: 5 Aug 04 Posts: 172 Credit: 4,023,611 RAC: 0 |
> well I think it's not quite every hour, it depends on your user preferences > for each project, i.e. if you set CPDN up as a "resource share" of 100 and > SETI as 100 they will share equally. I hope it's no less than an hour, > because if people have "leave suspended apps in memory" to false that will hit > a lot of lost checkpoint work (a CPDN checkpoint is every 144 timesteps, which > can be 10-15 minutes or longer between checkpoints depending on computer > speed). > As far as I can tell, about once an hour, BOINC makes a decision of what to run on that CPU next. I have 3 processes setup 5:1:1, and the minimum run seems to be an hour for the two that are set to 1, the the one that is set to 5 seems to get longer chunks. I think 2 and 3 hours. jm7 |
Send message Joined: 5 Aug 04 Posts: 250 Credit: 93,274 RAC: 0 |
> I hope it's no less than an hour, > because if people have "leave suspended apps in memory" to false that will hit > a lot of lost checkpoint work (a CPDN checkpoint is every 144 timesteps, which > can be 10-15 minutes or longer between checkpoints depending on computer > speed). > Isn't it better to set that value standard to Yes in the preferences then? It's now standard at No. Or at least, it was for me. -------------------- Jordâ„¢ <img src="http://boinc.mundayweb.com/cpdn/stats.php?userID=2&trans=off"> |
Send message Joined: 8 Aug 04 Posts: 5 Credit: 234,470 RAC: 0 |
> Isn't it better to set that value standard to Yes in the preferences then? > It's now standard at No. Or at least, it was for me. I would assume they don't want to TAKE resources by default. They would rather have the user willing GIVE up those resources. There are less lost steps if you select yes, but you'd rather not force people to give up memory space. <img src="http://boinc.mundayweb.com/seti2/stats.php?userID=1017"> <img src="http://boinc.mundayweb.com/predictor/stats.php?userID=115"> <img src="http://boinc.mundayweb.com/cpdn/stats.php?userID=20"> |
Send message Joined: 5 Aug 04 Posts: 907 Credit: 299,864 RAC: 0 |
I did have it set to default to "leave in memory" but David A told me the SETI/BOINC default is going to be "don't leave in memory", and since these are global prefs I wanted to be consistent with other projects. |
Send message Joined: 5 Aug 04 Posts: 250 Credit: 93,274 RAC: 0 |
Yes, but hello, the Seti 4.03 client when crunching takes up only 13MB. Between the two clients for CPDN there's about 55MB (I don't even count the 300MB of virtual memory it takes). Also considering a Seti unit will take up to 8 hours (maybe) max, whereas the CPDN unit takes 30 days, wouldn't you want to build in a safety margin for those people who don't yet know how BOINC works? I just mean that a Seti unit going flunky after 50% of crunching isn't so bad: download a new one. Would you want this to happen on a system that runs the climate model? Also a little warning for the people that don't use seperate prefs per project would be of use. Maybe that there's going to be a project which doesn't want the leaving in memory option (like Seti, although Seti runs great with the option on ;)), so people change their prefs but forget that if you change it in one project, it changes for all projects running under that location. -------------------- Jordâ„¢ <img src="http://boinc.mundayweb.com/cpdn/stats.php?userID=2&trans=off"> |
Send message Joined: 5 Aug 04 Posts: 39 Credit: 14,887 RAC: 0 |
Hopefully, at some point, the preemptive scheduler will take checkpointing into account when stopping WUs. How to do this hasn't been fully planned yet but it surely sounds like a good idea for everyone. |
Send message Joined: 5 Aug 04 Posts: 5 Credit: 24,702 RAC: 0 |
The BOINC website has a bit on this, the url is http://boinc.berkeley.edu/cpu_sched.php "Starting with version 4.00, the BOINC core client does time-slicing. This means that the core client may switch back and forth between results of different projects. This is done in a way that allocates CPU time according to the 'resource shares' you have assigned to each project." |
©2024 climateprediction.net