Mean CPU-time for S6Bucket tasks
13 May 2011 9:58:20 UTC
Topic 195788
(moderation:
Hallo !
Yesterday I got my first S6Bucket file. It will take some days to crunch all the files loaded more early, but I´m interested to know how long these new files may run.
From the Server Status Page I learned, that there where spend 38 CPU weeks within the last week, but we are much shorter than this, and did crunch successfully 9,101 WU within the same time rage. I suggest that this CPU weeks are the sum of the CPU time used to crunch all of the tasks within this time window. So 38 CPU weeks are 38*7*24=6,384 CPU hours. Dividing this by 2[tasks/WU]*9101[WU], because there are always 2 tasks per WU needed, I get 0.351[h/task]. That seems to me much to short. But what is wrong and how long do they really run ?
Kind regards
Martin
Mean CPU-time for S6Bucket tasks
)
Hi
the 9000 WUs are still in progress! There ca. 9000 tasks to send and 9000 taks in progress.
On my laptop with core i5 and HT on the tasks take 27500s. The S5 tasks took 30000s. This is only a 8% difference in runtime.
RE: the 9000 WUs are still
)
If you look on the Server Staus page in the field "S6Bucket search progress" you will see just now, there are allready done 9879 files. Together with the 46 CPU weeks listed above under Cumputing/S6Bucket, you get 0.391[h/tasks], which is about the same as before, but far away from reallity. But why?
Taking your crunching time, I get for 9879 Wus 898 CPU weeks.
Thanks for your answer.
Kind regards
Martin
RE: If you look on the
)
This is just the number of WUS that were generated into tasks. The more accurate number of processed tasks or WUs is the "task valid" number which is now at 346.
It think the CPU weeks are only calculated with the validated tasks [(CPU weeks
(from successful tasks last week)]. Last week the tasks for the S5-run were generated in one day and the tasks to send went up to 400000. With your theory this would have an huge effect on the CPU weeks for the S5 run but it didn´t. It is still less than the pulsar search.
If you calculate the runtime for the S6 task with the "task valid" number you get a more realistic number. >10h/task
You're lucky to get me right
)
You're lucky to get me right now because even though I did check that this application in my preferences, I do not get ....
Here are some of the run
)
Here are some of the run times from these WUs on my i7 870 with hyperthreading on.
Here's some stats on this
)
Here's some stats on this hyperthreaded i7 machine for its last 50 ( fifty ) Gravitational Wave S6 GC search v1.01 (SSE2) work units :
Run Time ( ie. wall clock ) -> average of 28640.6 seconds with standard deviation of 729.9 seconds
CPU Time ( ie. thread time) -> average of 24550.3 seconds with standard deviation of 161.0 seconds
{ Average CPU/RUN = 0.858 }
Cheers, Mike.
( edit ) Bear in mind the CPU is also servicing two NVidia cards for CUDA jobs.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal