| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: OpenZika
Thread: super low credit for some super long Zika WU |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 44
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Sadly, a week on the high variability continues to be a feeder for killing credit... Zika avg runtime at it's lowest since start, credit at lowest I'm not helping on the Zika credit at the moment: I've just switched from HST back to Zika and one machine started at 2.5 credits per hour, still ever so slowly increasing. |
||
|
|
SekeRob
Master Cruncher Joined: Jan 7, 2013 Post Count: 2741 Status: Offline |
The [additional] killer in the system it seems is, that stats are also maintained at host level. So going out at 5 hours for X credit and then when the lot has gravitated down, come back in, but then jobs having an entirely different duration, 1 hour. Current credit 20, yours did 5 hours before, now 1 hour, so whamma, 4 credit [That's kind of the picture I'm having of this crapaud].
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
A couple of strange low claimed credit cases on a machine running only Zika, where my _2 units are repairs of an _1 Error. These claims of 6.1 and 7.0 are in contrast to normal claims around 50 for this sort of CPU time on this machine (including a normal result that reported in between these 2 abnormal ones). It appears as though the existence of a wingman returning Error caused my low claims. Weird.
ZIKA_ 000024581_ x1yks_ YFV_ NS3hlcs_ 0479_ 2-- Microsoft Windows 10 Professional x64 Edition, (10.00.10586.00) 705 Pending Validation 19/06/16 09:23:10 20/06/16 21:47:40 1.45 6.1 / 0.0 ZIKA_ 000024581_ x1yks_ YFV_ NS3hlcs_ 0479_ 1-- Microsoft Windows 7 Enterprise x64 Edition, Service Pack 1, (06.01.7601.00) 705 Error 19/06/16 06:33:13 19/06/16 09:22:50 0.00 46.8 / 0.0 ZIKA_ 000024581_ x1yks_ YFV_ NS3hlcs_ 0479_ 0-- Microsoft Windows 10 Professional x64 Edition, (10.00.10586.00) - In Progress 19/06/16 06:32:55 29/06/16 06:32:55 0.00 0.0 / 0.0 ZIKA_ 000024581_ x1yks_ YFV_ NS3hlcs_ 0048_ 2-- Microsoft Windows 10 Professional x64 Edition, (10.00.10586.00) 705 Pending Validation 19/06/16 09:23:10 20/06/16 22:40:13 1.63 7.0 / 0.0 ZIKA_ 000024581_ x1yks_ YFV_ NS3hlcs_ 0048_ 1-- Microsoft Windows 7 Enterprise x64 Edition, Service Pack 1, (06.01.7601.00) 705 Error 19/06/16 06:33:13 19/06/16 09:22:50 0.00 46.8 / 0.0 ZIKA_ 000024581_ x1yks_ YFV_ NS3hlcs_ 0048_ 0-- Microsoft Windows 10 Professional x64 Edition, (10.00.10586.00) - In Progress 19/06/16 06:32:55 29/06/16 06:32:55 0.00 0.0 / 0.0 |
||
|
|
SekeRob
Master Cruncher Joined: Jan 7, 2013 Post Count: 2741 Status: Offline |
Circumstantial, but with the new v7.08 science app, it's probably the credit stats got booted [up]. Previous week the average ran ~25cr/hr average, and yesterday it jumped to 27.4cr/hr [checked numbers twice]. If it is in 10 days still like that, a hurray may be in order, but I'm a doubting Thomas on that token.
cr=credit, is what you see on the Result Status pages o!O |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges: 
|
Before deciding about outlying I think the least should be to check the number of jobs that the WU had to compute, or better to correctly set set the GFlops size according to the number of jobs.
----------------------------------------I have just finished a WU with 165 jobs instead of the usual 2x ones. Of course it needed 13 CPU hours instead of the usual 1.5 but it claimed (and got) half less credits than the usual small ones. I will not quit WCG just for that after so many years of similar discrepancies, but I can understand that too many contributors find such behavior frustrating and shocking, and do leave. Edit: This big WU was sized exactly the same as the small ones. ---------------------------------------- [Edit 1 times, last edit by JmBoullier at Jul 9, 2016 2:14:36 AM] |
||
|
|
SekeRob
Master Cruncher Joined: Jan 7, 2013 Post Count: 2741 Status: Offline |
Tech said that all the individual ligand runtimes [there are many variations] would be known afetr a few months, and thus when packaging them, the number zipped in a job would be having fairly even runtime, target 2 hours average. That part of 'average' appears to be achieved, with a current running mean of 2.09 hours.
----------------------------------------Anyway, the second half of yesterday caused the full day average credit/hr to end up as 34.43, off my chart as the y-axis tops at 33. :O))) (Saw similar happening when HST1 got a new app a week or 2 ago, now tailing off again, but still high at 31.18/Hr). [ot] Glad you're still around, Jean. (Hopefully not waiting on P3 of HCMD, as that will be a long wait)./ot [Edit 1 times, last edit by SekeRob* at Jul 9, 2016 7:24:28 AM] |
||
|
|
Crystal Pellet
Veteran Cruncher Joined: May 21, 2008 Post Count: 1408 Status: Offline Project Badges: 
|
Extreme long running task with Number of tasks = 189. Why are the jobs inside the task called tasks here where in the other sub-projects the sub-tasks are called jobs? In-consequent.
And of course low claiming credit: ZIKA_ 000035333_ x3mla_ Anthrax_ BaNadDApochB_ 0386_ 0-- Linux 3.2.0-105-generic 708 Pending Validation 7/7/16 17:55:22 7/8/16 22:52:56 16.79 44.0 / 0.0 |
||
|
|
SekeRob
Master Cruncher Joined: Jan 7, 2013 Post Count: 2741 Status: Offline |
Asked the question before at various times. Absent reply, wild guess
jobs = when there's dependencies within the task tasks = if a [variable] number of independent tasks are packaged for efficiency |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges:
|
Rob, I am not worried that the techs cannot precisely estimate WUs, it's obviously not easy.
----------------------------------------And in fact their current average of about 1000 GFlops per job is not far from reality for most "short" WUs. What puzzles me is why they don't simply use this average when they create WUs, i.e. why they don't simply weigh a 165-job WU to 165,000 GFlops instead of the 19,xxx that they arbitrarily put in almost every WU ? Right now I have another medium-big WU with 83 jobs and it has been estimated at 19,009 "as usual" while 83,000 would be close to what it will have used when completed. Of course such a so-simple trick would probably not satisfy Dr A.'s love for sophistication (complexification ?), but it's not Dr A. who runs WCG as far as I know?  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
This was another biggy, 14 hours to crunch 190 tasks. Normal Zika jobs run around 2 hours on this machine. Claimed credit is same per hour as normal ones, though.
ZIKA_ 000035552_ x3mla_ Anthrax_ BaNadD_ chA_ A_ 0426_ 1-- Microsoft Windows 10 Core x64 Edition, (10.00.10586.00) 708 Pending Validation 08/07/16 06:43:12 11/07/16 06:47:33 14.12 465.6 / 0.0 |
||
|
|
|