The last time I looked, there were only a handful of simulations available.
This is not really enough to see any problems.
A good simulation takes several hours to run.
If there were a few hundred simulations available it would take a month to
run through them all to see results. This looks like a good BOINC project.
The proposed scheduler could be submitted, and a set of BOINC clients could
grind through all of the simulations in a day or so. The analysis could
then take place to see if it fixed the proposed problem, and whether it
made any other problems worse.
jm7
David Anderson
<[email protected]
ey.edu> To
[email protected]
01/08/2010 01:20 cc
PM BOINC Developers Mailing List
<[email protected]>
Subject
Re: [boinc_dev] [boinc_alpha] EDF
and multi CPU tasks.
I don't understand what you mean.
Simulation lets you study the interactions of
long/short jobs, multithread jobs, GPU jobs, multiple projects, etc.,
quickly and with quantitative and reproducible results.
Anything involving the real BOINC client doesn't have these properties.
[email protected] wrote:
> It would be good if we had a suite of tests to run CPU schedulers through
> and a proposed CPU scheduler change could be tested quickly. I proposed
> setting this up as the "BOINC" project tasks. The memorex in this case
> could get us answers about the effectiveness much faster than live does.
> The problem is that the edge conditions do not happen frequently. Most
of
> the time on most machines any CPU scheduler will work. It is only
> sometimes that there is a major problem where some task will be reported
> weeks late, or a resource is ignored for a day or two...
>
_______________________________________________
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.
