On Mon, Sep 21, 2009 at 12:27:12PM -0700, David Anderson wrote:
> The current direction is described here (design doc in progress):
> http://boinc.berkeley.edu/trac/wiki/AutoFlops
> The idea is to make things as simple as possible for projects,
> while still getting accurate job completion estimates and fair credit.
The main problem is that you assume there is a big main loop that can
be counted, but some times there simply isn't. There are generic "it can
compute everything" type applications (math, chemistry, physics,
meteorology...) where the input is basically a very high level
scientific programming language that the application interprets. I fail
to see how a boinc_app_units()-like interface could be used in this
case. IMHO for such applications the only entity who can give any kind
of performance estimation is the person/program that generates the work
units.
Gabor
--
---------------------------------------------------------
MTA SZTAKI Computer and Automation Research Institute
Hungarian Academy of Sciences
---------------------------------------------------------
_______________________________________________
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.
