| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: BOINC Agent Support
Thread: priority of the WU |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 7
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hi,
In the queue of the WU for my platform I have one with a deadline much before the others. Question: is it possible to manually change the order of the priority for WU? Boinc is doing it automatically? If not, according to the queue will be processed first the WU with late deadline, which is not the best theoretically. Thanks, L. |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
The processing follows the FIFO rule which order is interrupted if jobs are in deadline trouble. As soon as BOINC determines that your short deadline task needs processing ahead of the FIFO rule it will put it front of the queue so it will finish 24 hours before deadline.
----------------------------------------Yes, you can manually manage the order of processing by suspending other tasks, but that is often the beginning of trouble developing. Just let BOINC do what it's much better at... scheduling jobs.
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at Aug 5, 2009 7:10:25 AM] |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges: 
|
The processing follows the FIFO rule which order is interrupted if jobs are in deadline trouble. As soon as BOINC determines that your short deadline task needs processing ahead of the FIFO rule it will put it front of the queue so it will finish 24 hours before deadline. The "finish 24 hour-before-deadline"-rule was removed in v5.10.xx, you're thinking about v5.8.xx and earlier clients. Instead, BOINC will finish tasks "Connect..." + "Switch between applications..." before the deadline. So, with a default "Connect..." of 0.1 days and "Switch..." of 1 hour, high-priority-tasks can finish so close as 3.5 - 4 hours before deadline.  "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Sadly, more of the "change of behavior without general notice". Well, don't ask me why 45 minute jobs that have an 8 day deadline, with a 4 hour project switch time are pushed ahead of the queue into HP, time and again for the same project within minutes of arrival (Now tested 4 non WCG projects doing this). Has 6.6.39 alpha a fix for that switching and queue jumping to high priority [blunt] BS. Not wanting to hijack the thread though, so maybe resume the topic I started several weeks ago at Berkeley, then ball dropped and a run made for the more important (so were the comments of the admin). Proper and reliable scheduling seems to me one of the most important BOINC features without grossly overusing the Project Weight assigned. A project that has consistent 45 minute jobs, yet sits with 100,000 negative LTD and rising and keeps pulling work unless manually stopped.
----------------------------------------On the topic of changes, wonder if developing a version/feature/rule matrix would help. Then support people can quickly check without the oblicatory 1 month on the ball observation per version study. Users could then quickly check how their version is supposed to operate. Maybe a volunteer who evidently very well knows when what functions, you maybe ;>) Sorry, we don't have the ultimate evil smoking smiley here as it gets me steaming
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges:
|
Sadly, more of the "change of behavior without general notice". Well, don't ask me why 45 minute jobs that have an 8 day deadline, with a 4 hour project switch time are pushed ahead of the queue into HP, time and again for the same project within minutes of arrival (Now tested 4 non WCG projects doing this). Has 6.6.39 alpha a fix for that switching and queue jumping to high priority [blunt] BS. Well, has been chasing badges lately so haven't really had the time to put v6.6.xx through extensive testing yet, but have still seen how low-resource-share projects now runs "high priority" while before they waited until just enough time to run "high priority" to be returned before deadline... Not sure if it's correct, but it can look like v6.6.xx does something like this: estimated_run_time = estimated_run_time(v5.10.xx) / runnable_resource_share So, if project-#1 has 1-hour-task, 8-day-deadline and resource-share 1, and project-#2 has resource-share 999, you'll get: estimated_run_time = 1 hour / 0.001 = 1000 hours. And of course, 1000 hours > 8 days => deadline-trouble => runs HP. So, seems this to fit your behaviour, or... ? Not wanting to hijack the thread though, so maybe resume the topic I started several weeks ago at Berkeley, then ball dropped and a run made for the more important (so were the comments of the admin). The release of "Progress Thru Processors" did have a deadline, so short-term a re-shuffle of priorities was need. But, the changes done seems to be mostly setup-related, and many of these was anyway planned for next release. Also, they have done some bug-fixes to v6.6.xx in parallell, even haven't released a new alpha-build yet to see if any more bugs is present. Proper and reliable scheduling seems to me one of the most important BOINC features without grossly overusing the Project Weight assigned. A project that has consistent 45 minute jobs, yet sits with 100,000 negative LTD and rising and keeps pulling work unless manually stopped. v6.6.xx always tries to keep "Connect..." filled with work, and if the zero LTD-projects either doesn't give any work or doesn't give enough work to completely fill the "Connect...", other projects is immediately tried (if contactable). In older clients the large negative LTD-projects isn't tried before after some time. Would guess complaints from dialup-users running out of work too early is behind this particular scheduling-change... The behaviour of "Additional..." is also changed, atleast my experience is v6.6.xx only tries to fill "Additional..." with negative-LTD in case of idle cpu. Meaning, if "Connect..." is zero, low-resource-share projects will ony download work in cases the client would otherwise sit idle. On the topic of changes, wonder if developing a version/feature/rule matrix would help. Then support people can quickly check without the oblicatory 1 month on the ball observation per version study. Users could then quickly check how their version is supposed to operate. Maybe a volunteer who evidently very well knows when what functions, you maybe ;>) Sorry, we don't have the ultimate evil smoking smiley here as it gets me steaming Hmm, a feature-matrix would be a nice idea... ... but writing FAQ's seems to be your job...  "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Very unpaid same as your technical history posts... so are you up for the challenge? If it needs a pretty plz, you'll have it now, sitting dog, paw up ... I'll put them in the FAQ's with embossed credit (grin)
----------------------------------------
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges:
|
Very unpaid same as your technical history posts... so are you up for the challenge? If it needs a pretty plz, you'll have it now, sitting dog, paw up ... I'll put them in the FAQ's with embossed credit (grin) Well, just some major highlights would be: v4.05: Removed the 2nd. cache-setting, instead filled-up to 2x cache-setting. v4.20: Stopped multiplying cache-setting with 2. v4.35 (April 2005): Stopped running solely based on resource-share and FIFO. Client made aware of deadlines and added "high priority". v4.45: Client sends a list of all work, so re-issue "lost" work and so on can work. v5.2.xx: too fuzzy recollection, but hopefully got basic code to stop asking for more work then cache full with more work than could handle before deadline. Also split BOINC Manager away from client. v5.2.6: Applications can report FLOPS and IOPS instead of relying on the BOINC-benchmark. v5.3.19: Actually the release-build for BBC/CPDN, uses libcurl instead of older network-code. v5.4.xx: Added long_term_debt and so on, to better try folliwing the resource-shares. Support for Account Managers. v5.8.xx (January 2007): Major re-writes, many rules of how to ask for more work, and then to stop asking for more. Example, like not then a task suspended, then too many uploads, deferring downloads, too large cache-size and so on. v5.8.17 (only non-windows): Due to a bug, server-aborting of unneeded tasks doesn't work correctly before in this and later builds. v5.10.xx (July 2007 and onwards): Added 64-bit-support. (hmm, wasn't the simple GUI also introduced in v5.10.xx?) Re-inroduced a second cache-setting. Removed the "finish 24 hours before deadline"-safety-margin, removed the "if 'Connect..." larger than / 2 to deadline, stop asking for work". v5.10.14: For reporting, removed the "report 'connect...' after uploaded", add instead "report 24 hours after uploaded", "report 'connect...' before deadline", add option 'report_results_immediately'. v6.2.xx Not aware of any scheduling-changes. Separates data-directory from program-directory for better Vista-support. Adds "sandbox" for better security. v6.4.xx: Changes to run CUDA, not aware of much else. v6.6.xx (March 2009): Many changes to handle cpu, CUDA and SMP from same project, project using fractional cpu. "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
|