| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: Suggestions / Feedback
Thread: Get work only on completion |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 7
|
|
| Author |
|
|
ttjph
Cruncher Joined: Jun 23, 2010 Post Count: 2 Status: Offline Project Badges: 
|
Hi,
Long time cruncher, first time poster - sorry if this is already covered elsewhere. I run BOINC/WCG on a system with a pretty good per-thread speed but which is probably only on for an hour a day, on average. A typical MCM workunit (for example) completes in about four hours - which is a significant fraction of the seven-day 'shelf life'. I have my workunit cache set to 0.001 days (and I've also tried zero), but even so the system downloads new workunits days before the previous ones complete - the result being that many workunits exceed their deadline, unless I micro-manage (which frequently involves aborting units immediately after download). I've tried leaving BOINC to 'learn' my system's usage pattern, but if this is having any effect, it's very slow and effectively imperceptible, and results in many hours of lost crunching time in the meantime. Given that I have a fast, reliable, always-on internet connection, would it be possible either to add an explicit option to only download a new workunit when the previous one has completed (i.e., a CPU core is now idle); or for "zero" in the cache settings to mean zero? Thank you! |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Depends on the client, but v7 setting of MinBuf / MaxBuf to zero only fetches new works minutes before the current jobs are completed... this is how I run my old very occasionally on Centrino duo.
----------------------------------------There's a MinWork config setting but think that only fetches 1 single workunit at the time, meaning 1 core works, the rest are idle. Edit, yes: <fetch_minimal_work>0|1</fetch_minimal_work> Fetch one job per device (see --fetch_minimal_work). --fetch_minimal_work Fetch only 1 job per device (CPU, GPU). Used with --exit_when_idle, the client will process one job per device, then exit. New in 6.11 In context, a device is a CPU or a GPU. [Edit 2 times, last edit by Former Member at Jun 20, 2015 1:54:06 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
If your machine is on just 1-2 hours a day, you may consider 'hibernation' at end of use rather than shutting down [see other forum posts on this]. Next time you power the computer it's very quick to start, your applications are up as you left them [web browser/ doc programs etc] and BOINC resumes exactly where it left off rather than going back to previous checkpoint. This will save lots of computing time in a week.
----------------------------------------[Edit 1 times, last edit by Former Member at Jun 20, 2015 1:59:38 PM] |
||
|
|
ttjph
Cruncher Joined: Jun 23, 2010 Post Count: 2 Status: Offline Project Badges:
|
Thanks - the pointer to minbuf/maxbuf did it once I'd found where they were.
I've previously only using the web-interface profile settings - I'd assumed that they would be master if I didn't touch the local options, but clearly there are different options locally and they override at least the cache setting in the profile. Seems to be working exactly as I wanted now. :-) |
||
|
|
BobCat13
Senior Cruncher Joined: Oct 29, 2005 Post Count: 295 Status: Offline Project Badges: 
|
If you are only running WCG, and using Boinc client 6.10.30 or higher, you could set the resource share (Project weight at WCG) to 0. That would request a minimal amount of work no matter what the minbuf/maxbuf is set to.
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Thanks - the pointer to minbuf/maxbuf did it once I'd found where they were. I've previously only using the web-interface profile settings - I'd assumed that they would be master if I didn't touch the local options, but clearly there are different options locally and they override at least the cache setting in the profile. Seems to be working exactly as I wanted now. :-) If only once you visited the local prefs and took OK instead of cancelling out, the web prefs of that moment + whatever you may / may not have changed locally becomes locked in as local override file. As of that moment the web prefs for that device no work no more [except a few such as project selection and project weight]. Hit the Clear button in the local screen and your device reverts to using the web preferences again, the reason you did not got what you wanted before. The 7.6 client has an improved interface, telling who's preferences are in effect [local or which profile of the project, who's lastly were updated] An extensive transition explanation I wrote in 2012 of old/new: https://secure.worldcommunitygrid.org/forums/wcg/viewthread_thread,32530 The descriptions on the website are still those of older clients. What to do if volunteers use versions in the range from 5.4 to 7.6 ;? |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Sample of the now tested [still alpha, still not public beta] 7.6.3 preferences screen. Total reorg, most all descriptions speaking for themselves, and little pop-up texts when hovering an input field for more explanation.
 |
||
|
|
|