I believe I have figured out the problem with the saturation numbers being
off.

jmcleod2

3730085     Goldbach's Conjecture Project 1/31/2011 11:11:55 AM   [rr_sim]
405186.46: wu_2011-01-21215224-45197_0 finishes after 5185.73
(42.14G/0.01G)
3730086     Goldbach's Conjecture Project 1/31/2011 11:11:55 AM   [rr_sim]
wu_2011-01-21215224-45197_0 misses deadline by 196708.15

The task finishes about 5,000 seconds from now, but since it is already
past its computation deadline, it is registering as saturating the cpu by
nearly 200,000 seconds.

There are a couple more problems with using rr_sim to calculate saturation.

1)  rr_sim is appropriately for CPU scheduling using a maximum value for
run time length.  Unfortunately, this will tend to underfetch work as the
mean and median run times are typically less than the maximum run time.
This conflict cannot be fixed without a separate calculation for work
fetch.
2)  rr_sim uses the resource share for the next project to get a work fetch
to increase the apparent run times.  There are two problems with this.  It
is the wrong measure, and it needs to be different for CPU scheduling and
work fetch.  CPU scheduling needs the extra safety margin, but work fetch
does not.  Anyway, the correct measure is the number of hours of work that
is likely to be downloaded / the expected time to deadline.

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