| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
| World Community Grid Forums
|
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 12
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
In all four of my BOINC clients, all running under Win10 Pro with latest updates, when a new task starts, it usually isn't the one that's due the soonest.
There seems to be a preference for running projects other than MCM, and only when a bunch of those are nearing deadline will those start instead of tasks that have more time until deadline. This can be a problem especially on my laptop, where I will either have it off during travel or suspended while working. Tasks that run out of Deadline order may wind up not having enough time to finish, thus wasting the CPU cycles spent on trying to get there. What I've been doing is manually suspending tasks so that the only tasks that can run are the ones closest to deadline, but surely the obviousness of wanting to run tasks in deadline order is apparent. There also seems to be a recent problem with rebooting a computer to shake loose a stuck task so it can complete. When I suspend tasks so that the ones closest to the deadline can run, this leaves tasks partially complete and in a Waiting To Run state. However, when I return from a reboot lately, I've been getting Computation errors on some of them. Just a little bit ago, which prompted me to take the time to pursue this, four tasks all popped up with Computation error immediately after rebooting. So - the primary question is this: How do I get the BOINC client to start tasks in Deadline order automatically so that those with the closest deadline run before those due later? |
||
|
|
Crystal Pellet
Veteran Cruncher Joined: May 21, 2008 Post Count: 1410 Status: Offline Project Badges:
|
So - the primary question is this: How do I get the BOINC client to start tasks in Deadline order automatically so that those with the closest deadline run before those due later? First set in your Device Profiles for the project limits a maximum number of tasks you want to have in buffer. Then set in BOINC Manager in Computing Preferences at the Computing tab to 'Store at least ... days of work' to 10. |
||
|
|
ca05065
Senior Cruncher Joined: Dec 4, 2007 Post Count: 328 Status: Offline Project Badges:
|
The normal running state for BOINC is First In First Out (FIFO). If it calculates that some work units will not be completed by their deadline it will change to Earliest Deadline First (EDL). It is not possible to change this method of working.
I usually run my computer for 12 hours per day using a queue length of about one day as the longest unscheduled outage I have ever seen from WCG is less than 12 hours. BOINC then looks after itself quite happily. The problem of computation error of work units after a restart mostly applies for me on Windows for Microbiome Immunity Project - It occurs very rarely on Linux Mint. I reduce the impact of the restart computation errors by hibernating my PC instead of shutting it down. However Windows or some other software forces the need for a restart every few days. Hibernating the PC also prevents the loss of computation time since the previous checkpoint which can be up to 30 minutes per work unit for MIB work units. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Thanks. I left the # of tasks at unlimited for now, but did change the number of days to 10 on the first device I'll test. It immediately suspended tasks running out of deadline order and fired up the ones due earlier.
I'll send WCG a request to implement an option that directly addresses this. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Thanks for this information. Yes, it was the MIP projects that got all the computation errors on this last reboot.
Are you saying that hibernating the PC will shake the frozen task loose? That this is an alternative to rebooting to get it to finish? |
||
|
|
KerSamson
Master Cruncher Switzerland Joined: Jan 29, 2007 Post Count: 1684 Status: Offline Project Badges:
|
Since WCG is very reliable by distributing WUs, I do not see any necessity to maintain a 10-day buffer. You should reduce to maybe 0.5 or 1 day additional buffer. If your buffer has a limited size, you should not experience any trouble by achieving the deadline.
----------------------------------------Yes ! MIP causes some troubles and is demanding. Cheers, Yves |
||
|
|
Crystal Pellet
Veteran Cruncher Joined: May 21, 2008 Post Count: 1410 Status: Offline Project Badges:
|
Thanks. I left the # of tasks at unlimited for now, but did change the number of days to 10 on the first device I'll test. It immediately suspended tasks running out of deadline order and fired up the ones due earlier. Not a good idea not to limit the number of tasks. With at least 10 days of work you will flooding your machine with tasks. That's why I told you to limit the number of tasks in your device profile. @KerSamson: My reply was to give a method on his primary need to run the tasks EDF and not FIFO. |
||
|
|
ca05065
Senior Cruncher Joined: Dec 4, 2007 Post Count: 328 Status: Offline Project Badges:
|
@areteone
Hibernating a PC is a method of avoiding the MIP computation errors and avoiding the loss of computation time starting from the previous checkpoint when re-booting a PC. My understanding of the process is that the OS writes the contents of memory to disk before shutting down. On startup, the contents of memory on disk are read back in, and processing is resumed as if the hibernation had not occurred. |
||
|
|
hchc
Veteran Cruncher USA Joined: Aug 15, 2006 Post Count: 865 Status: Offline Project Badges:
|
I used to think EDL was preferable to FIFO, but then if you have projects with differing deadlines -- say, 7 days and 10 days -- then the 7 day project will always run, and the 10 day project will never run.
----------------------------------------The best solution I've seen is to let the client do its thing (FIFO) but reduce the buffer to something reasonable like 0.5 or 1 day. This way the entire buffer is worked in FIFO order, and deadline is never a problem.
[Edit 1 times, last edit by hchc at Aug 27, 2019 9:56:03 PM] |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges:
|
I used to think EDL was preferable to FIFO, but then if you have projects with differing deadlines -- say, 7 days and 10 days -- then the 7 day project will always run, and the 10 day project will never run. EDL uses the true deadline which is computed by adding the allowed number of days for computation** to the distribution date.All jobs will be computed in time when their deadline comes closer to the current day. Except, maybe, for people who use much too large a buffer. . ** Yes we often call it deadline but this is improper. ![]() Edit: By the way, I fully agree to your best solution. ---------------------------------------- [Edit 3 times, last edit by JmBoullier at Aug 28, 2019 3:35:46 AM] |
||
|
|
|