| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: Help Fight Childhood Cancer Project Forum
Thread: MrKermit |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 44
|
|
| Author |
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Hi MrK,
----------------------------------------In your particular case, send a mail to support@worldcommunitygrid.org. They'll be able to help on making sure that all completed work is reported and all tasks not completed get aborted. Generally, I'd say, set additional days low, to have least jobs queued and connect time to 0.0 and "always connected". Indeed, you may have caused a little reverberation in the FAAH cycle where there were reports of "nothing but repair jobs" shortly after your "emergency". On that, I know someone who's doing rendering services on hundreds of cores and saves the clients all-in away and puts them back whatever device becomes available, on a dynamic basis. The expiration/lifetime of jobs is normally 10 days, repair work 4 days. If considered, it might be a question for the techs if there is a way to block repair jobs yet still be a quick returner, which is the premise for getting repair job assignments. Certainly when you have to step away, a whole lot of hours of computed time on work in progress would not be lost if you manage to get back before that... else the jobs would simply expire and be reissued when due. [in a nutshell and will poll the techs to have a lookin on the whitepaper you seek for your situation] edit: Yes, Jean below remembered how to block repair jobs... set the project switch time high.
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at Oct 1, 2009 9:45:41 AM] |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges: 
|
Since I think you are "WCG-only" a quick and easy way to not receive short deadline jobs is to set the "Switch between applications" parameter to a value >= 4 days. For example 6,000 minutes will do.
----------------------------------------And keeping the "cache extra work" value at half a day or even less should leave your devices enough time to process WUs in time when they have a break or when they are not doing business at full speed.  |
||
|
|
nasher
Veteran Cruncher USA Joined: Dec 2, 2005 Post Count: 1423 Status: Offline Project Badges: 
|
happy to see you hoping in when you can.
----------------------------------------I do see that you managed to get 20+days done yesterday hopefully you wont recieve too many emergencies but eithor way just haveing whatever computers you can have run for however long is impressive myself i manage 3-6 days per day... not super impressive but it gets me there... burst crunching or sustained cruncing both have there benifits... yes you may not be able to run the project more than a few hours per month but in those hours you will burst ahead of many of the sustained crunchers... I am sure there may be ways to sneek in a little more work or make it easyer to reload or whatever you need to do to streamline the process... good luck and always glad to see anyone crunching  |
||
|
|
Movieman
Veteran Cruncher Joined: Sep 9, 2006 Post Count: 1042 Status: Offline |
Movieman, We have ~ 600 Intel Harperton 2.8Ghz, Dual socket/Quad Core machines with 8GB ram, and ~300 Intel Woodcrest 3.0Ghz Dual socket/Dual Core Machines with 8GB ram. All the systems are in maintenance mode at the moment, so I thought I'd kick it up a notch over here while we wait on a few fixes from misc. vendors  We'll be largely out of skewing the WCG numbers by the end of the month.In total run time we aren't as competitive, but in Results and Points I think we are above average, so pretty good efficiency node for node. Still in all, on a cost per WU I would bet most of the other teams out there would look much more efficient. It's like showing up to the Solar Races with a Charged up Tesla and acting like you're in the race fair and square :) We're glad we can help a bunch of good causes and put UIUC in the limelight for a bit. If some of our larger plans work out this year we may be able to help more soon. Cheers! Dear Lord! 600 Harpers and 300 Woodcrests?  And here I am in my little "pond" with 2 harpers(3710 and 3415mhz) one Gainestown at 3458, a Yorkie at 4171 and an old tech dual Sossaman at 2000mhz..and I thought I had a bunch in the house.. Can't imagine 900 dualies.. Sounds like a handfull to keep up and running. All my best to you and good luck. PS:Set up ONE Gainestown rig, you will fall over when you see what it does in terms of output vs the harpertowns and the upcoming Gulftowns will absolutely destroy the Gainestowns in terms of output. I'm hearing 22% better clock for clock and less heat per core as they are 32nm vs the Gainestowns 45nm. Picture a 12 core dualie running at 4500mhz doing 24 threads but take your heart pills first.   |
||
|
|
|