By Client Estimated credit, I assume you mean something like wall time *
device flops * C? Could we benchmark each device on the system so that if
it is assigned to a particular device we will have some idea how that
device performs? For example if a CPU has a rating of 1000 and GPU #1 has
a rating of 5000 and GPU#2 has a rating of 15000, it would be less than
useful to lump the two GPUs together. If a task is assigned to GPU #1 it
will run at 1/3 the speed that it would if it was assigned to GPU #2, and
it just seems to be a better solution to have two distinct numbers for
this.
Yes, CPUs on a particular host should be all the same, but GPUs are often
different.
I would still like to see servers report a list of supported device types.
This would make a consistent work fetch possible for the faster devices
(from my scenario with 3 projects with equal resource shares, only one of
which can use the GPU).
jm7
David Anderson
<[email protected]
ey.edu> To
<[email protected]>
10/27/2010 02:41 cc
PM BOINC Developers Mailing List
<[email protected]>
Subject
Re: [boinc_dev] proposed scheduling
policy changes
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
_______________________________________________
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.
