Cancel previous msg - I don't see a bug after all.
If anyone wants to clarify or otherwise decrease the stupidity
of the code, feel free.
-- DPA
Nicolás Alvarez wrote:
> El Martes 07 Jul 2009 14:27:26 David Anderson escribió:
>> The weighting is actually quadratic; i.e., when the job is half done
>> the estimate is 75% from the dynamic and 25% from the static.
>>
>> However, I think I see a serious bug in the code (see below).
>> Anyone else see it?
>
> Last time I looked at this code, I thought something quite stupid was being
> done. Then I analyzed it more carefully and it turned out to be right, just
> written in a confusing way.
>
> I'll try a "second pass".
>
>> // Returns the estimated CPU time to completion (in seconds) of this task.
>> // Compute this as a weighted average of estimates based on
>> // 1) the workunit's flops count
>> // 2) the current reported CPU time and fraction done
>> //
>> double ACTIVE_TASK::est_time_to_completion(bool for_work_fetch) {
>> if (fraction_done >= 1) return 0;
>> double wu_est = result->estimated_duration(for_work_fetch);
>> if (fraction_done <= 0) return wu_est;
>> double frac_est = (elapsed_time / fraction_done) - elapsed_time;
>> double fraction_left = 1-fraction_done;
>> double wu_weight = fraction_left * fraction_left;
>> double fd_weight = 1 - wu_weight;
>> double x = fd_weight*frac_est + wu_weight*fraction_left*wu_est;
>> return x;
>> }
>>
>> double RESULT::estimated_duration(bool) {
>> return
>> estimated_duration_uncorrected()*project->duration_correction_factor; }
>>
>> double RESULT::estimated_duration_uncorrected() {
>> return wup->rsc_fpops_est/avp->flops;
>> }
>> _______________________________________________
>> 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.
