| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: Discovering Dengue Drugs - Together - Phase 2 Forum
Thread: DDDT-2 Work Unit distribution updated |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 121
|
|
| Author |
|
|
HutchNYC
Advanced Cruncher United States Joined: Nov 27, 2005 Post Count: 97 Status: Offline Project Badges: 
|
The main benefit of 7 day reply is that a member cannot scoop up more than 7 days of work (per core) at once, so the scheduler will get to spread a wider net to more computers. I understand what you're saying and it won't affect me either way. I crunch 24/7 and one of the Type-A WU's ran for over 51 hours on my i7-920 with no interruptions or resume from a prior checkpoint. Just thinking about what happens when a "normal" cruncher gets one of these on a computer that they only use for a few hours a day or even using it during a normal business day for 8 hours. Long periods of crunching time could be lost with a checkpoint only happening every hour or so. So then you potentially end up with someone who has had their computer crunching the WU for 40 hours in a week, still won't make the 7-day deadline, and has nothing to show for their 40 hours crunching contribution. Just something to consider is all I'm saying.  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Just thinking about what happens when a "normal" cruncher gets one of these on a computer that they only use for a few hours a day or even using it during a normal business day for 8 hours. I believe that pattern makes the machine 'not reliable' so they would not be eligible if the WU are being released only to reliable systems. |
||
|
|
jasm580
Senior Cruncher USA Joined: Dec 20, 2007 Post Count: 157 Status: Offline Project Badges: 
|
They would get a "Won't finish in time" message
----------------------------------------
-Jasm
|
||
|
|
Mathilde2006
Senior Cruncher Germany Joined: Sep 30, 2006 Post Count: 269 Status: Offline Project Badges: 
|
They would get a "Won't finish in time" message My problem is, that the A-Wus are coming with incorrect ETA-time. On my computer they start with 45-53 hours, but they end up all with more than 70. Maybe it would better, to start them with 7 days timebomb, but after completed distribution to the clients the timebomb should be moved to a higher level at a later date. Lets say one or two days before the 7 days deadline will be reached. The WUs should be crunched with priority. Workaround for the 'too late' problem. Badge hunters can't overcache.  |
||
|
|
X-Files 27
Senior Cruncher Canada Joined: May 21, 2007 Post Count: 391 Status: Offline Project Badges:
|
1. New type A work units will be sent to reliable hosts only with a 7 day turnaround. This means they will get scheduled first on the grid to the generally faster 24/7 machines on the grid. And add a quota per device - Lets say number of processors * 2. Like gpugrid which needs fast turnaround.   |
||
|
|
Jack007
Master Cruncher CANADA Joined: Feb 25, 2005 Post Count: 1604 Status: Offline Project Badges: 
|
And add a quota per device - Lets say number of processors * 2. Is that with Hyperthreading?   |
||
|
|
darth_vader
Veteran Cruncher A galaxy far, far away... Joined: Jul 13, 2005 Post Count: 514 Status: Offline Project Badges: 
|
And add a quota per device - Lets say number of processors * 2. Is that with Hyperthreading? I don't think BOINC can tell the difference, so yes. In this project you can really see the downside of hyperthreading since the run times are so long. All 4 of my WUs ran 33-35 hours on a 2.5 GHz C2D. Yet the much faster I7s are taking longer as was mentioned a few posts ago: I crunch 24/7 and one of the Type-A WU's ran for over 51 hours on my i7-920 with no interruptions or resume from a prior checkpoint. - D |
||
|
|
Zanth
Advanced Cruncher USA Joined: Aug 18, 2008 Post Count: 88 Status: Offline Project Badges:
|
This sounds good... provided it helps one of my very reliable machines get some work. :)
---------------------------------------- |
||
|
|
Jack007
Master Cruncher CANADA Joined: Feb 25, 2005 Post Count: 1604 Status: Offline Project Badges:
|
And add a quota per device - Lets say number of processors * 2. Is that with Hyperthreading? I don't think BOINC can tell the difference, so yes. In this project you can really see the downside of hyperthreading since the run times are so long. All 4 of my WUs ran 33-35 hours on a 2.5 GHz C2D. Yet the much faster I7s are taking longer as was mentioned a few posts ago: I crunch 24/7 and one of the Type-A WU's ran for over 51 hours on my i7-920 with no interruptions or resume from a prior checkpoint. - D Actually, my Q6600 took less time doing 4 than my I7 did 8, by several hours. Since my I7 picked up a Retread I suspended 4 out of 8 tasks and it's 97.7% done in 32.5 hours (50 hours or so with 8 cores running). Trying to get Dengue back as quick as i can. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
While I agree with the objective behind a 7 day deadline, I think in reality it would make things even worse overall. From just the first short release of the initial 1000 Type "A's", nearly 25% weren't returned in time with a 14 day deadline. I think cutting the 14 day deadline in half would only compound the issue. I reckon it'll be much better. As discussed in another thread, with the 14-day deadline, machines could be sitting working on some other project, grab a couple of type As, then go back to the other project for a week, then finally start on the type As, then find out that the time estimate is wrong and fail to finish on time. If the type As have a 7-day deadline, the client will process them sooner. I suspect that if 25% aren't back in time on a 14-day deadline, then that might increase a little with a 7-day deadline, but those which do complete being twice as fast will more than make up for it. |
||
|
|
|