| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: BOINC Agent Support
Thread: Question about BOINC scheduling within project |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 4
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
By "project" I mean WCG as a whole, not one of its research projects.
What determines the next wu to be run *within a project*? All these years I thought and observed that wu's within the same project were run in order of earliest deadline, unless some cpu time had already accumulated on a later wu. Then last night my host received a nice batch of DDD-T2 work with a 7 day deadline. Today I see that its "marking time" cache of HCC workunits, all with later deadlines and shorter estimated duration, are being crunched first. Nothing is in deadline-crunch mode. I had to suspend the HCCs in order to get the wu's returned faster -- with the annoying side effect of preventing any more project work from being retrieved meanwhile. Are project wus actually run in order of received date? Not wanting to take the time to hunt up and ask on a BOINC dev forum, just wondering what you all have seen. Thanks. |
||
|
|
rilian
Veteran Cruncher Ukraine - we rule! Joined: Jun 17, 2007 Post Count: 1460 Status: Offline Project Badges: 
|
Scheduler algorith is changed in BOINC 6.x
----------------------------------------First-in-first-out scheme does not work ... |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
The host is on 6.6 and yes, fifo seems dumb.
|
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges: 
|
The host is on 6.6 and yes, fifo seems dumb. BOINC has always ran fifo within a project, except if there's a risk of missing the deadline. In case of hitting memory-preference-limit it can also run in a different order, and atleast in some versions all GPU-work runs in EDF-mode. Now, part of the reason to run fifo has to do with multi-project-support, since anyone running 50/50 with CPDN and another project will definitely not want to have the CPDN-task stuck for nearly a year in cache while only the other project runs EDF... But, even if only a single project, running fifo normally has it's advantages over running only EDF. Example, a dual-core computer running a single project with 2 wu-sizes, one "large" with 30 days deadline and 3 days cpu-time, and a "small" with 2 days deadline and 1 hour cpu-time. If this computer gets a single "large" wu, and a long range of "small" wu, if this computer runs strictly EDF, the "large" wu will sit in the cache with zero progress for 28 days, until it's only 2 days to the deadline, and finally this wu has now the earliest deadline and start running... Not surprisingly, a task that uses 3 days cpu-time and start to run 2 days before deadline has zero chance of being returned by the deadline... Oh, and even if BOINC does consider the deadline, and will start running the "large" wu 3.1 days before deadline is over, there's still multiple disadvantages to running EDF compared to FIFO. Among these is, if "wrong" estimate (something that is very common), the "large" wu can take longer to run than estimated, and if is delayed to close to the deadline it has no hope of finishing before deadline. Another thing to consider is, most projects doesn't have 100% uptime, so by delaying until very close to deadline, it's bigger chance for having connection-problems, either due to project being down, or you've got more local problems...  "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
|