On 27/10/10 19:41, David Anderson wrote:
> 
> On 27-Oct-2010 11:38 AM, [email protected] wrote:
> 
>> The proposed work fetch policy will also mean that more work will be
>> fetched from the projects where the machine is least effective, assuming
>> FLOP counting is used to grant credits.
> 
> That's a good point.
> I'm now leaning towards client-estimated credit
> rather than actual credit as a basis.
> (Also because of the issues that Richard raised)
> 
> -- David

Isn't that going to penalise projects that optimise their applications
to use host time more efficiently?


Controlling share by RAC (client estimate, or server reported) will
remove all incentives to optimise applications because the scheduler
will balance out any optimisations to give more resource *time* to
non-optimised applications to (wastefully) boost their RAC up to share...

Would a more robust way to go be to schedule by resource time,
individually for each resource unit?


Scheduling using resource time used, optimised applications can then
fairly get more work done. For a later step, resources could be
allocated to make best use of the available resources, for example when
there are both GPU and CPU versions of an application available...

Regards,
Martin

-- 
--------------------
Martin Lomas
m_boincdev ml1 co uk.ddSPAM.dd
--------------------
_______________________________________________
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