> The only mechanism for changing it is to modify the duration correction
> factor.  The duration correction factor will correct itself over time
> anyway, but it is possible to set it by hand to be closer to where it will
> eventually stabilize.
>
> jm7

The problem with relying on DCF is that no adjustment to DCF is made until a 
task completes, even if the initial estimate is drastically low - so all 
calculations which rely on DCF, like work fetch, do not start to 
self-correct until it's too late.

A case in point: I have been in extended conversation with the administrator 
of the AQUA project. They are issuing test workunits of varying duration, 
but haven't got the hang of adjusting <rsc_fpops_est> to ensure BOINC has 
something to base its estimates on. I'm seeing the same estimate for 
everything from 8-!M (15 minutes) to 240-5M (280 hours). The latter equates 
to a DCF of around 200, which is going to be a problem: but I have a host 
running one of these where the DCF is still below 8 from earlier, smaller 
tests, and presumably the DCF won't be corrected for the best part of a 
fortnight.

Of course, it would be better if all BOINC projects made at least a 
reasonable attempt at adjusting their work estimates, but not everyone does 
this - and I've singularly failed to persuade AQUA to join the ranks of the 
ones that do. I believe David has been in touch with this project - perhaps 
you could have a word?

[ Does anyone run training courses for new BOINC administrators? ]

But back to DCF - if DCF could be edged upwards incrementally once the 
elapsed runtime has exceeded the initial estimate, it would help to avoid 
excessive work fetch on unruly projects. 


_______________________________________________
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