OK, I can agree with that.  I would still like to add the exception for
idle CPU though.  That would make it:

if a project has wacky DCF, ask it for work
only if the number of devices used by runnable jobs
is less than (resource share)*ninstances, OR a CPU is idle and it is
otherwise the best candidate.

jm7


                                                                           
             David Anderson                                                
             <[email protected]                                             
             ey.edu>                                                    To 
                                       [email protected]              
             01/15/2010 12:24                                           cc 
             PM                        [email protected]          
                                                                   Subject 
                                       Re: Work fetch tweak.               
                                                                           
                                                                           
                                                                           
                                                                           
                                                                           
                                                                           




I think that's too extreme;
if project A has a wacky DCF and several other projects don't,
project A will never have more than 1 job running,
which may violate resource share.

The following policy might be better:
if a project has wacky DCF, ask it for work
only if the number of devices used by runnable jobs
is less than (resource share)*ninstances

-- David

[email protected] wrote:
> ===================================================================
...
> Could we change this to be:
>
> Only fetch work in this case if that project has no other tasks on the
> system, or there is an IDLE CPU?
>
> 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