Idle CPUs should not going to be a part of the design for small queues.
Especially since the default queue is 0.
jm7
Martin
<[email protected]
o.uk> To
Sent by: [email protected]
boinc_dev-bounces cc
@ssl.berkeley.edu
Subject
Re: [boinc_dev] 6.6.20 and work
04/28/2009 08:41 scheduling
AM
TarotApprentice wrote:
>> Date: Mon, 27 Apr 2009 15:42:26 -0400 From: [email protected]
[...]
>> We need to run the check for what should be run on a variety of
>> events and at least once every task switch interval.
>>
>> We need to do schedule enforcement on a different set of events,
>> one of which is going to occur very frequently - checkpoint.
>>
>> jm7
>
> Which begs the question if we DIDN'T do it at every checkpoint (or
> time) but rather based upon TSI would things improve? I don't know
Indeed, is there *any* need for testing the schedule any more often than
min(TSI, minimum WU deadline, 0.75*(cache length time) ) ?
Note that CPUs will go idle for very small cache sizes if the transfer
times are long. This should be accepted! If a user wants a very small
cache, let them have that as set. Perhaps issue a warning that the cache
size is smaller than the expected WU request/transfer time.
For example, I have some very small memory devices that could help out
for small WUs and zero cache.
If a download pulls in WUs that spoil the show, then that is a fault of
code elsewhere in delivering such unpalatable WUs. On the client we can
still afford (to wait) to find out at the next scheduled scheduling
test. EDF or whatever then sorts out whatever mess as best it can.
Perhaps automatically junk hopeless WUs if they have not been started?
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.
_______________________________________________
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.
