David Anderson wrote on 08/12/2010 22:35:
> http://boinc.berkeley.edu/trac/wiki/PrefsRemodel
>
> This isn't something we're going to do anytime soon,
> but we may as well discuss it.
That page asks if anyone uses the following.
* CPU scheduling period
Yes. One malariacontrol.net application recently had a set of WUs which
ran for longer than usual. The longer the task ran, the slower it
became in starting to make progress after being scheduled. I had a task
which took 19 minutes to start progressing the last time it was
scheduled (limited non-BOINC activity, lots of disk access during the
"dead" time and reported times of 7:31:42 elapsed and 6:28:06 CPU).
Increasing the CPU scheduling period to 7 hours allowed those tasks to
complete in a single period.
* leave apps in memory (should default to No)
Yes. A number of projects have applications which perform their
checkpoints at fixed points. The most extreme example I've seen is
WCG's CEP2, which only checkpoints at the end of each discrete job; I
recently had a task which took 7 hours to run one of those jobs.
Without LAIM that task could have been pre-empted and removed from
memory with a lot of lost CPU time.
* Checkpoint interval
I have it set to 10 minutes. Some applications write a lot of data to
disk when they checkpoint; the malariacontrol.net application I
mentioned was writing more than 250MB per checkpoint.
--
Ian
_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.
