Welcome to MilkyWay@home

Extreme (inaccurate) time estimates...?

Message boards : Number crunching : Extreme (inaccurate) time estimates...?
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile NullCoding*
Avatar

Send message
Joined: 23 Sep 10
Posts: 24
Credit: 58,711,243
RAC: 0
Message 42514 - Posted: 3 Oct 2010, 8:00:56 UTC

Although I'm somewhat new to MW and BOINC in general, I know enough about the concept of CPU crunching to know that something is a little off here:



No, it doesn't actually take that long. Earlier today I had a task reading 1108:20:00 that had run for an hour already and another (on PrimeGrid) reading 00:25:50 that had not started...

Am I doing something wrong, or is it just an error in communication between my CPU and the app? For reference, it's a 2.53GHz i5 on OSX 10.6.4; my GPU is incompatible and the computer runs 24/7. There doesn't seem to be an issue with reporting tasks, but measuring and estimating them (for resource distribution, etc) seems to be affected - for example, I can't for the life of me remember setting any preferences labeling certain tasks as "high priority."

Help a newbie? :)
ID: 42514 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Matt Arsenault
Volunteer moderator
Project developer
Project tester
Project scientist

Send message
Joined: 8 May 10
Posts: 576
Credit: 15,979,383
RAC: 0
Message 42515 - Posted: 3 Oct 2010, 8:24:10 UTC - in response to Message 42514.  

The nbody time estimates are very wrong.

The time one of the nbody workunits varies widely depending on the parameters. Some will take only a few minutes, some may take days. Tne initial estimates you see should be very high compared to what they actually will be. Before the estimate was too low, so after a long time running BOINC was killing them, so the initial estimate was just raised to be very high. I've been sort of coming up with a better initial prediction which eventually will help with the initial time estimate.
ID: 42515 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile NullCoding*
Avatar

Send message
Joined: 23 Sep 10
Posts: 24
Credit: 58,711,243
RAC: 0
Message 42530 - Posted: 3 Oct 2010, 19:58:10 UTC

Ok, thanks.

No task has ever taken more than 2 days (as the machine runs 24/7) but the 200+ hour estimates threw me a bit - thought I had set up my preferences wrong or something.

Luckily I've not had a problem with BOINC killing tasks that "time out;" maybe this is why.

So what does "high priority" mean, exactly? is that something I set somehow, or is it the WU that calls itself that?
ID: 42530 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Matt Arsenault
Volunteer moderator
Project developer
Project tester
Project scientist

Send message
Joined: 8 May 10
Posts: 576
Credit: 15,979,383
RAC: 0
Message 42541 - Posted: 4 Oct 2010, 0:14:29 UTC - in response to Message 42530.  

So what does "high priority" mean, exactly? is that something I set somehow, or is it the WU that calls itself that?


I'm not sure how BOINC decides to prioritize workunits. I think it's a combination of workunit deadlines, and preferences for what percentage of the time to devote to whatever projects you're connected to.
ID: 42541 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
DanNeely

Send message
Joined: 6 Oct 09
Posts: 39
Credit: 78,881,405
RAC: 0
Message 42545 - Posted: 4 Oct 2010, 4:07:17 UTC

In normal use boinc keeps track of short and long term deviations from the allocated work distributions (short and long term debt in clientstate.xml) and uses them to balance the distribution of time between various projects.

High priority mode kicks on whenever boinc thinks a work unit is in danger of not being completed in time at its the current resource share. Whenever that happens it goes into earliest deadline first mode which is indicated by high priority in the status column and allows debt levels among the various projects to build up.
ID: 42545 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
bill

Send message
Joined: 15 Jul 09
Posts: 12
Credit: 45,145,989
RAC: 0
Message 42546 - Posted: 4 Oct 2010, 4:34:48 UTC

I'd like to know why BOINC will run work units that aren't due for days ahead
of work units that are due much sooner on the same project; sometimes even running the units with further in the future due dates high priority rather than the work units due in the near future.
ID: 42546 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile NullCoding*
Avatar

Send message
Joined: 23 Sep 10
Posts: 24
Credit: 58,711,243
RAC: 0
Message 42554 - Posted: 4 Oct 2010, 14:59:38 UTC

Aha...thanks DanNeely et. al...I thought it was something to do with the way I use GridRepublic to manage things (although I'm not entirely sure I do it right).

Resource-share doesn't necessarily affect prioritizing, then? I imagine it would, if BOINC doesn't feel I'm allocating enough CPU time to MW@H, whose tasks take the longest of the five projects I run.

Seems like I've don't get WUs that're due in more than a week, save for the really big ones from MW 0.31 and a few Sierpinski Prime Problems....
ID: 42554 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : Extreme (inaccurate) time estimates...?

©2024 Astroinformatics Group