Re: [boinc_dev] [boinc_alpha] Card Gflops in BOINC 6.10

[Raistmer]

> I think it's feasible to maintain these estimates dynamically,
> based on actual job runtimes.
If app share CPU time with another app (for example, some app that uses 
partially CPU and partially GPU) BOINC estimates will be completely screwed.
I see it right now running SETI MB CPU on SETI main and AP CPU+GPU on SETI beta 
on quad.
Those MB tasks who happened to be on the same core as AP tasks exhibit 
completion times ~9 hours while other ~1-1,5 hours.
When such "long playing"tasks completed BOINC estimates for SETI jobs just go 
mad. Then, after completion of "normal" task BOINC tries to reduce its estimate 
but makes it much slower than rising it again.
If these estimates will be used for crediting instead of FLOPs , well, credits 
will be even more "funny" area than they are right now...
_______________________________________________
boinc_dev mailing list
boinc_dev@ssl.berkeley.edu
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.