Better estimate of amount of work for w/u's
log in

Advanced search

Questions and Answers : Wish list : Better estimate of amount of work for w/u's

Author Message
TheSeb
Send message
Joined: 29 Aug 12
Posts: 2
Credit: 7,273
RAC: 0
Message 278 - Posted: 7 Oct 2012, 21:12:59 UTC

I've been crunching work units for a while now for FMAH, and I've noticed that the work units tend to be of very different sizes. This is probably related to the complexity of the docking (amount of docking points?).

Is there any way of estimating the amount of work required for a particular work unit? Now each work unit is estimated to take 1799 GFLOPS of work, though some tasks take over 20 minutes to finish, and others finish within 30 seconds. This means that the BOINC estimate of the time required to finish a work unit also fluctuates wildly.

The problem I experience with this is that when the BOINC estimate is high it retrieves only a few work units which may finish in a matter of minutes, while at other times BOINC retrieves more than a hundred work units when the estimate is low. As I can't let BOINC run all the time (only when my laptop is plugged in), it sometimes takes several days to process all FMAH tasks.

I've sort-of fixed this by only requesting work for a maximum of 0.2 days, but if you could improve this situation by giving a better estimate of the size of each work unit I'd appreciate it.

Ant Chubb
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 18 Jul 12
Posts: 118
Credit: 1,019,875
RAC: 0
Message 283 - Posted: 10 Oct 2012, 9:01:58 UTC - in response to Message 278.

I'm afraid that estimating the time to completion is really tricky with this project. Each compound being tested has a different number of rotatable bonds, and then the search space it finds itself in is different for each receptor. As we're searching 19,000 compounds against 2,500 receptors, and each docking time is dependent on the kind of CPU used, estimating the time will be pretty complicated!
If Kevin figures out some clever trick I'm sure he'll implement it.
Either way, we appreciate the work done by all the volunteers, and have a long window for each job to complete, so it should not be an issue if the work takes days to get back to us. Or am I missing something here?

TheSeb
Send message
Joined: 29 Aug 12
Posts: 2
Credit: 7,273
RAC: 0
Message 290 - Posted: 11 Oct 2012, 18:21:59 UTC

Thanks for your response.
It is not a major problem for me, though every time BOINC downloaded >100 large FMAH-workunits, it had to work through these first before WUs of other projects were downloaded again, which is why I asked whether a better estimate could be provided.

Requesting work for only 0.2 days proves to be a good work-around, so if estimating the time for each WU is difficult to implement it's not a problem for me.

Questions and Answers : Wish list : Better estimate of amount of work for w/u's


Main page · Your account · Message boards


Copyright © 2017 Dr Anthony Chubb