| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: BOINC Agent Support
Thread: Why BOINC not execute near expiring WUs first? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 4
|
|
| Author |
|
|
farta
Cruncher Joined: Aug 2, 2009 Post Count: 7 Status: Offline Project Badges: 
|
I'm caching WUs more than default. I'm aware that algorithm for sorting and picking up WUs from cache is inexplicable. Right now it is about 9/10/2015 11:00 local time, 4 tasks with ETA between 9/17 & 9/20 are running, 23 WUs are in cache queue, and 4 tasks with ETA 9/13 seems at the end of the queue.
I'm worrying about if it would cause WU expire. Is there an option about priority algorithm (which I may changed by mistake)? Or is there a flaw in BOINC's algorithm? Or is there nothing to be worried about, BOINIC would guarantee WU won't expire due to not executing them in time? |
|||
|
||||
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Not quite inexplicable, but close
---------------------------------------- . The basis of scheduling is FIFO, First In First Out, no matter the deadline. Only when there's a deadline-exceed threat will jobs move in the EDF, Earliest Deadline First order, automatically (and some will debate this). BOINC has though some safeties build-in. For instance the shorter deadline tasks are moved ahead if half of their deadline has already been used up. If though there is so much cached that everything starts running in 'panic' mode, you're in trouble. This is why the rule of thumb is to never cache more than half of the shortest deadline project you participate in. E.g. if the MCM repairs are due in 2.8 days and your buffer is greater than 1.4 days worth of work, these MCM repairs will be scheduled ahead quite soon. A second safety is, that clients that do not return work within 48 hours will not get those repair / short deadline jobs. The [WCG] server keeps track of what to assign.In a nutshell [Edit 1 times, last edit by Former Member at Sep 10, 2015 7:37:37 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
BTW, the rescheduling algorithm runs ones per minute or 10 seconds, don't quite remember. It re-computes total estimated buffer runtime needed to process all tasks and then assesses if EDF is required for any task in the buffer.
|
||
|
|
farta
Cruncher Joined: Aug 2, 2009 Post Count: 7 Status: Offline Project Badges:
|
Not quite inexplicable, but close . The basis of scheduling is FIFO, First In First Out, no matter the deadline. Only when there's a deadline-exceed threat will jobs move in the EDF, Earliest Deadline First order, automatically (and some will debate this). BOINC has though some safeties build-in. For instance the shorter deadline tasks are moved ahead if half of their deadline has already been used up. If though there is so much cached that everything starts running in 'panic' mode, you're in trouble. This is why the rule of thumb is to never cache more than half of the shortest deadline project you participate in. E.g. if the MCM repairs are due in 2.8 days and your buffer is greater than 1.4 days worth of work, these MCM repairs will be scheduled ahead quite soon. A second safety is, that clients that do not return work within 48 hours will not get those repair / short deadline jobs. The [WCG] server keeps track of what to assign.In a nutshell Thanks a lot. I'm running FAAH & MCM, caching 0.5 day of task. I guess it would be OK.  |
||
|
|
|