I agree with your reasoning about the delays or even potential complete failure 
of current credit-granting mechanisms, David.

But I feel the statement about Resource Share being expressed in Flops is more 
of an assertion than an axiom, and that the volunteer's perspective should be 
considered too. I'd best leave it at that.



>________________________________
> From: William Stilte <[email protected]>
>To: David Anderson <[email protected]> 
>Cc: "[email protected]" <[email protected]> 
>Sent: Thursday, 12 June 2014, 18:08
>Subject: Re: [boinc_dev] EXIT_TIME_LIMIT_EXCEEDED (sorry, yes me again, but 
>please read)
> 
>
>Considering that we do not count flops but time...
>
>Retreating to semantics?
>
>And I've got a few cents to add on using RAC but I've run out of time to
>share that particular balloon popper.
>
>
>2014-06-12 18:45 GMT+02:00 David Anderson <[email protected]>:
>
>> To reiterate (from discussions several years ago):
>>
>> - Resource shares apply to a host's total FLOPS, not time on particular
>> devices.
>>   This is the appropriate semantics.
>>
>> - We chose not to use RAC as a basis for scheduling because
>>   a) there can be long and unpredictable delays in granting credit,
>>      because of replication
>>   b) jobs may not be granted credit for a variety of reasons,
>>      in which case the client would (undesirably)
>>      keep running jobs for that project.
>>
>> -- D
>>
>> _______________________________________________
>> 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.
>
>
>
_______________________________________________
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