I think that these would be would be good next steps. On my currently running system the debts are just growing outwards from 0...
After several hours of running (a reset happened about 0700): 12/9/2009 10:13:47 AM n...@home [std_debug] CPU STD delta -4.41 (0.09*127.70 - 15.77) 12/9/2009 10:13:47 AM PrimeGrid [std_debug] CPU STD delta -12.93 (0.02*127.70 - 15.77) 12/9/2009 10:13:47 AM World Community Grid [std_debug] CPU STD delta 18.92 (0.89*127.70 - 94.59) 12/9/2009 10:13:47 AM n...@home [std_debug] CPU STD -4107.29 12/9/2009 10:13:47 AM PrimeGrid [std_debug] CPU STD -15976.80 12/9/2009 10:13:47 AM World Community Grid [std_debug] CPU STD 20084.09 12/9/2009 10:13:47 AM Collatz Conjecture [std_debug] ATI GPU STD delta -15.77 (0.50*31.53 - 31.53) 12/9/2009 10:13:47 AM milky...@home [std_debug] ATI GPU STD delta 15.77 (0.50*31.53 - 0.00) 12/9/2009 10:13:47 AM Collatz Conjecture [std_debug] ATI GPU STD -3237.08 12/9/2009 10:13:47 AM milky...@home [std_debug] ATI GPU STD 3237.08 Note taht I have switched from doing MW to doing COllatz which will last about a day or so till the FIFO queue is empty. On Dec 9, 2009, at 5:22 AM, [email protected] wrote: > > A couple of ideas on STD. > > 1) Do not do the rebalance to have the mean be 0. The current rebalance > tends to disallow low resource share projects from the round robin - they > always seem to run in EDF. > 2) If a project does not have work, have the STD creep towards 0. If the > STD were to go towards 0 based on some fraction of wall time (either fixed > or based on resource fraction) it would tend to give projects that do not > have tasks for a while a normal start in the queue while giving projects > that get work immediately about the same position in the queue that they > would have had before. > > 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. _______________________________________________ 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.
