I have been thinking, and a temporary DCF might help with startup problems,
however, I believe it should be done a bit differently than the recent
experiment.

If a DCF is needed, make a temporary copy of the current DCF for each
project.

Add the current estimate for each task that is partly complete to the
temporary DCF using the same algorithm as is used to add completed tasks to
the DCF.  For projects where the fpops_est is extremely low, this will help
avoid some of the work fetch that occurs before the first task is
completed.  Later in life of the project, the impact is not likely to be as
large.

I would like to suggest that using the mean and standard deviation for DCF
might be more stable later in the project lifetime.  I would also suggest
that using the app version and fpops_est as keys to look up the current DCF
for a task would also be a good idea.  This would lead to a heirarchy of
DCF values to be used if a a lower level DCF is unknown.

If you have a DCF for a project, app version, fpops_est that is valid (for
statistics based, count > 1 is valid), use that.
else if you have a DCF for a project, app version that is valid, use that.
else if you have a DCF for a project, that is valid, use that.
else use the value 1.

This means that new projects would assume the value 1.  A new app would use
the project base DCF, and a new fpops_est would use the base for the app
version.

jm7

_______________________________________________
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.

Reply via email to