Advanced search

Message boards : Server and website : Question about credits after power failure

Author Message
Dave_In_Oz
Send message
Joined: 13 Jul 09
Posts: 32
Credit: 275,659,344
RAC: 127,411
Level
Asn
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 18508 - Posted: 3 Sep 2010 | 12:20:43 UTC

Hi,

I understand there are bonuses for returning tasks inside 24hrs and inside 48hrs. Owing to the outage, tasks completed inside 24hrs could not be returned and as a result got the lesser bonus. Will this be corrcted?

Profile skgiven
Volunteer moderator
Project tester
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,976,731,260
RAC: 128,946
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 18510 - Posted: 3 Sep 2010 | 13:27:49 UTC - in response to Message 18508.

The power outage did not prevent some of my systems from getting full credit, but it was a bit hit and miss which ones got full credit and which did not. My Fermi did for both tasks, as did two GT240's, but some GT240's did not and my GTX260 did not. All of my cards that missed the 50% bonus received the 25% bonus.

It's down to the Scientist to decide on the points, but I would be in favour of temporarily moving the 50% credit to 2days. Not sure how much work would be involved so cant even guess if it is realistic to ask for this.

Snow Crash
Send message
Joined: 4 Apr 09
Posts: 450
Credit: 539,316,349
RAC: 0
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 18513 - Posted: 3 Sep 2010 | 14:42:01 UTC

Seriously?
You would rather have them manage a special case temporary issue for bonus points due to a power outage they have no control over instead of doing real science?
____________
Thanks - Steve

Profile skgiven
Volunteer moderator
Project tester
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,976,731,260
RAC: 128,946
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 18515 - Posted: 3 Sep 2010 | 14:47:39 UTC - in response to Message 18513.

If it's a 5min task then yes, but not if it's half a days work.

Profile Saenger
Avatar
Send message
Joined: 20 Jul 08
Posts: 134
Credit: 4,396,501
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwatwat
Message 18517 - Posted: 3 Sep 2010 | 15:07:38 UTC

I "lost" a bonus as well, shit happens :)
So be it, don't bother too much with this, the real work an the data base stability is more important.
If it's just one or two clicks in a menu, do something, but otherwise: just let it be.
____________
Gruesse vom Saenger

For questions about Boinc look in the BOINC-Wiki

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1919
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 18533 - Posted: 5 Sep 2010 | 8:19:37 UTC - in response to Message 18517.

To return the credits individually would be impossible. We can increase the deadline for a while to receive for credits, but really it's just a temporary measure.

We are rather working on the new app which is already 5% faster and hopefully a bit more on the GTX460. We cannot check it as we don't have any.

gdf

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2023
Credit: 14,666,833,719
RAC: 1,029,309
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 18534 - Posted: 5 Sep 2010 | 8:58:12 UTC - in response to Message 18533.
Last modified: 5 Sep 2010 | 9:00:46 UTC

Working on a faster app would be better than spending time with any temporary measure.
The power outage affected everyone, so there is no need to compensate this.
We were lucky it didn't happened on friday night.
However, a bigger WU cache would be better for fast hosts running 24/7. Running out of work bothers me more than a lost bonus. (losing not just credit, but precious GPU time)

jjwhalen
Send message
Joined: 23 Nov 09
Posts: 29
Credit: 17,591,899
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 18538 - Posted: 5 Sep 2010 | 15:53:33 UTC - in response to Message 18534.

Working on a faster app would be better than spending time with any temporary measure.
The power outage affected everyone, so there is no need to compensate this.
We were lucky it didn't happened on friday night.
However, a bigger WU cache would be better for fast hosts running 24/7. Running out of work bothers me more than a lost bonus. (losing not just credit, but precious GPU time)


I'll agree with Retvari on this, point for point.

I understand the need for fast returns, results being used to build the next batch, etc.. But even a cache increase from 2 to 3 would help to at least partially cover future outages :)

Snow Crash
Send message
Joined: 4 Apr 09
Posts: 450
Credit: 539,316,349
RAC: 0
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 18539 - Posted: 5 Sep 2010 | 19:01:20 UTC - in response to Message 18538.

As soon as the minumum goes up to three WUs then there will a LOT more WUs that do not get returned within 48 hours and have to be resent. This not only causes more work for the project but the frequency of the complaints of "wasted" effort will increase. I would say if we could cache a minumum of two (which would be the same as we get today) and a max based on the average returned within 24 hours per gpu that might strike a reasonable balance between keeping our cards crunching and keeping the project on schedule. Yes, slow cards would only get two and the fastest cards could get about 10-12 max. Someone will post it is not fair that fast cards get more WUs but the reality is that slow cards are still caching relatively more time so doesn't that mean it is unfair to people who have fast cards?

Unfortunately any scheduling logic that is "non-standard" from a BOINC perspective causes the project to write their own custom code and also makes them have to review all BOINC server changes in greater detail to make sure they can roll over their custom code without breaking anything. And if we think it is difficult for us to deal with a power outage, think about how the really smart guys who are waiting to work with the results must feel!!!
____________
Thanks - Steve

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1104
Credit: 6,101,732,079
RAC: 11,685
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 18542 - Posted: 5 Sep 2010 | 21:08:33 UTC

I like the WU limit of 2 as long as the bonus limit is 1 day. This project doesn't have much downtime and BOINC now supports backup projects so setting Collatz for instance with a project priority of 0 will allow it to DL a minimal amount of work until GPUGRID returns. No wasted GPU cycles. Keep up the good work guys!

Profile skgiven
Volunteer moderator
Project tester
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,976,731,260
RAC: 128,946
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 18543 - Posted: 5 Sep 2010 | 22:00:23 UTC - in response to Message 18542.

Lots of really constructive points made about the credit system and potential outages. I’m sure the scientists will appreciate these being raised as it allows them to continuously think things over. The best balance is probably very difficult to find and fluctuates with different projects coming and going, and definitely with new technology. Ultimately it comes down to task productivity, so I hope the scientists can see through the various concerns and make the correct decision based on everything they know about the project (overall turnover optimization).
I think the ideal/easy situation would be that the GPUGrid team found a partner in a different country, so that if one base went down the other could act as a failover task reservoir. Perhaps someday they will be in a position to collaborate with another group, or one of the researchers will branch off/move on and establish a co-operative team in another establishment.

A word of caution about the long work units, these might start turning up only for Fermi cards. They are still rather long though, even for Fermis. Expect such tasks to take about 23h on a GTX460 and about 18h on a GTX470, roughly (depends on your OS, hardware and settings).
If you cache is high don’t expect full bonus points, unless the 50% bonus is pushed out to two days, until that mini-project finishes. If you are not bothered about the points just leave things as they are, and be happy that if there is an outage that will not bother you either.

Post to thread

Message boards : Server and website : Question about credits after power failure