do aborted tasks influence duration_correction_factor?

Message boards : Questions and problems : do aborted tasks influence duration_correction_factor?
Message board moderation

To post messages, you must log in.

AuthorMessage
borsti
Avatar

Send message
Joined: 18 May 06
Posts: 32
Germany
Message 37055 - Posted: 2 Mar 2011, 11:57:16 UTC

hi,

some time ago we've got some ATOM D510 devices and decided to give them a try on BOINC also (client 6.10.58, OS XP prof.).
In general they're working ok, but for WorldCommunityGrid the expected runtime is WAY out of bounds (e.g. calculated runtime 6 hours, in fact it took more than 2 days)!
When I notice some PC has got a WU which won't finish in time, I usually abort it.
Does BOINC take the partial crunching time in consideration for the DCF?

If yes, I don't need to do anything else, because with each aborted WU the estimation for the next ones will become better.
Otherwise I would have to either let them finish for no good, or manually correct the XML for each and every ATOM-PC... :(
cu/2
Torsten
ID: 37055 · Report as offensive
Profile Gundolf Jahn

Send message
Joined: 20 Dec 07
Posts: 1069
Germany
Message 37056 - Posted: 2 Mar 2011, 12:01:54 UTC - in response to Message 37055.  

How could do BOINC achieve that? To correct the DCF, it has to compare the estimated runtime with the real runtime. If you abort the task, there is nothing to compare with.

Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
ID: 37056 · Report as offensive
borsti
Avatar

Send message
Joined: 18 May 06
Posts: 32
Germany
Message 37059 - Posted: 2 Mar 2011, 16:32:02 UTC - in response to Message 37056.  

BOINC has some fraction?! (But of course I have to assume linear progress)

Example: the WU was calculated to finish in 10 hours.
After 1 hour 1% is done (both values are reported by the client!). So when I abort now, the estimated total runtime would be 100 hours. Since this is 10 times more than expected, the DCF should be raised accordingly.
cu/2
Torsten
ID: 37059 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15482
Netherlands
Message 37060 - Posted: 2 Mar 2011, 16:49:51 UTC - in response to Message 37059.  

The duration correction factor will only change at the end of the task. When you abort the task, the DCF is not affected, it'll stay at what it was. No matter what you think it should and should not do.

Now, WCG tasks do not have a deadline of 2 days, but one of 14 days. You want to tell me you can't finish them in that allotted time? For then your cache wishes may be too high.

I'm running BOINC only during 9pm - 7am during the week, and all hours in the weekend and manage to get all work for all 5 projects that I run out in time for their respective deadlines, including the Malaria project with their small deadlines and WCG/Einstein with their long run time and longer deadlines.
ID: 37060 · Report as offensive
borsti
Avatar

Send message
Joined: 18 May 06
Posts: 32
Germany
Message 37063 - Posted: 2 Mar 2011, 17:55:00 UTC - in response to Message 37060.  

When you abort the task, the DCF is not affected, it'll stay at what it was.

too sad. So I have to do manual corrections... :(

Now, WCG tasks do not have a deadline of 2 days, but one of 14 days.


This was just clarify the situation, not a "real life" example.
You want to tell me you can't finish them in that allotted time?


I think there's 2 factors influencing the somewhat bad behavior of the client here:

    - BOINC doesn't respect known working hours. When a computer is on 8 hours a day 5 days a week it calculates about 6 hours as a daily average. It will happily accept a WU on Friday with a deadline on Sunday...
    - WCG has not a high priority for me (about 20%)


So it happens quite often that I get a WU many days in advance, but BOINC prefers other projects. This would be fine if the expected runtime was ok; but now the WU sits around until the deadline gets critical, and then of course it's too late.

The only other idea I have would be to attach new ATOM's ONLY to WCG until the DFC has become reasonable (if there are no other projects, the WU's will be finished in time) and then attach the others. But it's likely to forget the 2nd step. ;)


cu/2
Torsten
ID: 37063 · Report as offensive

Message boards : Questions and problems : do aborted tasks influence duration_correction_factor?

Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.