Quota and credit calculations. Not bad, not bad at all.


log in

Advanced search

Message boards : Number crunching : Quota and credit calculations. Not bad, not bad at all.

Author Message
Sten-Arne
Volunteer tester
Send message
Joined: 1 Nov 08
Posts: 3513
Credit: 20,667,303
RAC: 21,874
Sweden
Message 1005177 - Posted: 17 Jun 2010, 9:54:11 UTC

This new quota stuff, and new credit calculation stuff, looks pretty promising.

The credit thing is a bit shaky at this early stage, with some WU's getting way low compared to before, and some getting higher than before. It's going to be interesting to follow this over time.

The quota thing, also looks promising. Over time it looks as if it is going to reward those that takes care of their WU's, and punishing those who deliberately abort them, or constantly turns in WU's that error out for one reason or another.

I'm looking forward to seeing over time how this develops.

____________

Profile Area 51
Avatar
Send message
Joined: 31 Jan 04
Posts: 965
Credit: 42,193,520
RAC: 0
United Kingdom
Message 1005179 - Posted: 17 Jun 2010, 10:16:36 UTC - in response to Message 1005177.


The quota thing, also looks promising. Over time it looks as if it is going to reward those that takes care of their WU's, and punishing those who deliberately abort them, or constantly turns in WU's that error out for one reason or another.


I agree that it looks promising, but I can't help feeling that the GPU crunchers are going to get hit when they generate -12 errors (something that can't be avoided).
____________

Richard HaselgroveProject donor
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8629
Credit: 51,435,784
RAC: 47,808
United Kingdom
Message 1005187 - Posted: 17 Jun 2010, 11:06:59 UTC - in response to Message 1005177.

It's going to be interesting to follow this over time.

I followed it over time: Beta message 39533.

Note that since that was posted, the number he first thought of has been doubled. So the intial spike should be even higher, and the long-term average should be lifted. But I expect the variability to remain.

Profile Francesco Forti
Avatar
Send message
Joined: 24 May 00
Posts: 281
Credit: 140,184,985
RAC: 48,257
Switzerland
Message 1005306 - Posted: 17 Jun 2010, 15:44:17 UTC - in response to Message 1005179.


The quota thing, also looks promising. Over time it looks as if it is going to reward those that takes care of their WU's, and punishing those who deliberately abort them, or constantly turns in WU's that error out for one reason or another.


I agree that it looks promising, but I can't help feeling that the GPU crunchers are going to get hit when they generate -12 errors (something that can't be avoided).

Even the autokill function should't penalize us.
Because now SETI knows the application we use (CPU or GPU) and must be able not to send VLAR to GPU. So we have not to kill them.
Who can agree?
FF

____________

Profile Area 51
Avatar
Send message
Joined: 31 Jan 04
Posts: 965
Credit: 42,193,520
RAC: 0
United Kingdom
Message 1005313 - Posted: 17 Jun 2010, 16:12:27 UTC - in response to Message 1005306.

Nice idea, but it doesn't work like that. I have received over 100 VLARs for my GPU today. Also, I think I am right in saying that the stock CUDA application can handle VLARs, so as far as SAH is concerned, there is no issue.
____________

Profile perryjay
Volunteer tester
Avatar
Send message
Joined: 20 Aug 02
Posts: 3377
Credit: 15,897,216
RAC: 10,980
United States
Message 1005316 - Posted: 17 Jun 2010, 16:17:44 UTC

Over time it looks as if it is going to reward those that takes care of their WU's, and punishing those who deliberately abort them, or constantly turns in WU's that error out for one reason or another.


I messed up a few days ago and lost a little over 100 WUs. They should start timing out over the next month and a half. The first of them on the 22nd of this month, the last of them on the 30th of July. By then everything should be running however they are trying to make it run so I guess these WUs will tell me what's what.

My apologies to my wingmen by the way but right after I lost them I got a new bunch to replace them so I really can't detach to clear them.
____________


PROUD MEMBER OF Team Starfire World BOINC

Sten-Arne
Volunteer tester
Send message
Joined: 1 Nov 08
Posts: 3513
Credit: 20,667,303
RAC: 21,874
Sweden
Message 1005340 - Posted: 17 Jun 2010, 17:58:43 UTC - in response to Message 1005187.

It's going to be interesting to follow this over time.

I followed it over time: Beta message 39533.

Note that since that was posted, the number he first thought of has been doubled. So the intial spike should be even higher, and the long-term average should be lifted. But I expect the variability to remain.


Hmm, well lots of variability there I'd say. We'll see how this gets reveived among the participants here. I predict a s***storm :-)

I'm not easily upset though, as long as the science we do is correct, I'm sitting down in the boat.


____________

Profile perryjay
Volunteer tester
Avatar
Send message
Joined: 20 Aug 02
Posts: 3377
Credit: 15,897,216
RAC: 10,980
United States
Message 1005344 - Posted: 17 Jun 2010, 18:19:29 UTC

Wonder when the storm from the members of other projects starts after they see some of us getting somewhere around 500 or 600 credits for one GPU WU? (I got one that paid 580 credits!}
____________


PROUD MEMBER OF Team Starfire World BOINC

Sten-Arne
Volunteer tester
Send message
Joined: 1 Nov 08
Posts: 3513
Credit: 20,667,303
RAC: 21,874
Sweden
Message 1005347 - Posted: 17 Jun 2010, 18:25:29 UTC - in response to Message 1005344.
Last modified: 17 Jun 2010, 18:26:14 UTC

Wonder when the storm from the members of other projects starts after they see some of us getting somewhere around 500 or 600 credits for one GPU WU? (I got one that paid 580 credits!}

Well, I've got some that normally would have give me over 100 and only gave 15-20. So I guess it average out somehow...
____________

gomeyer
Volunteer tester
Send message
Joined: 21 May 99
Posts: 488
Credit: 50,157,953
RAC: 0
United States
Message 1005350 - Posted: 17 Jun 2010, 18:36:14 UTC
Last modified: 17 Jun 2010, 18:37:04 UTC

I may have missed this in the plethora of opinion about the new scheduler setup, but my question is, now that the server is keeping track of CPU/GPU assignment what happens to work units that are reassigned using the rescheduler when they have been returned having been run by other than the platform to which they were originally assigned? Are they still valid?

Josef W. SegurProject donor
Volunteer developer
Volunteer tester
Send message
Joined: 30 Oct 99
Posts: 4299
Credit: 1,068,158
RAC: 1,012
United States
Message 1005352 - Posted: 17 Jun 2010, 18:57:24 UTC - in response to Message 1005350.

I may have missed this in the plethora of opinion about the new scheduler setup, but my question is, now that the server is keeping track of CPU/GPU assignment what happens to work units that are reassigned using the rescheduler when they have been returned having been run by other than the platform to which they were originally assigned? Are they still valid?

Still valid, yes. The statistics of work done and how long it took are probably applied to the application to which they were originally assigned, though. Reassigning VLARs to the CPU may not be much problem, the run time may be similar to what they would have been on GPU. General rebalancing could have far more impact.
Joe

gomeyer
Volunteer tester
Send message
Joined: 21 May 99
Posts: 488
Credit: 50,157,953
RAC: 0
United States
Message 1005356 - Posted: 17 Jun 2010, 19:18:49 UTC - in response to Message 1005352.

I may have missed this in the plethora of opinion about the new scheduler setup, but my question is, now that the server is keeping track of CPU/GPU assignment what happens to work units that are reassigned using the rescheduler when they have been returned having been run by other than the platform to which they were originally assigned? Are they still valid?

Still valid, yes. The statistics of work done and how long it took are probably applied to the application to which they were originally assigned, though. Reassigning VLARs to the CPU may not be much problem, the run time may be similar to what they would have been on GPU. General rebalancing could have far more impact.
Joe

Understood. Thanks Joe.

zoom314Project donor
Avatar
Send message
Joined: 30 Nov 03
Posts: 46499
Credit: 36,845,760
RAC: 5,046
United States
Message 1005357 - Posted: 17 Jun 2010, 19:25:42 UTC
Last modified: 17 Jun 2010, 20:18:24 UTC

Post moved elsewhere.
____________
My Facebook, War Commander, 2015

Sten-Arne
Volunteer tester
Send message
Joined: 1 Nov 08
Posts: 3513
Credit: 20,667,303
RAC: 21,874
Sweden
Message 1005489 - Posted: 18 Jun 2010, 2:36:35 UTC

So far seems to work as intended. Quota going up when returning valid work. Credit varying up and down, but may settle, or not :-)

I'm satisfied so far, and so are my computers.


____________

Message boards : Number crunching : Quota and credit calculations. Not bad, not bad at all.

Copyright © 2014 University of California