John, I know that you're a veteran BOINC user and very knowledgeable, but
are you running a variety of GPU projects, especially on machines with
multiple GPUs?  Work fetch with CPU projects is not as large a problem
unless you're trying to use an app_config.

Regards/Ed


On Wed, Feb 27, 2013 at 8:06 AM, McLeod, John <[email protected]> wrote:

> I believe you have missed part of the point of the work fetch algorithm.
>  It uses work fetch to balance out the work done over the long term to
> match your resource shares.  There is no guarantee that the client will be
> running exactly N of project A and M of project B.  In general, set the
> resource shares the way you want to share the work load, and don't try to
> micro manage it.  Just let it balance things out over the long term.
>
> It is possible to sign up for more projects than you have processors, and
> sometimes projects are down or are out of work.  Some projects have very
> tight deadlines, and need every processor when work is downloaded.
>
> -----Original Message-----
> From: [email protected] [mailto:
> [email protected]] On Behalf Of Ed A
> Sent: Tuesday, February 26, 2013 3:38 PM
> To: Rom Walton
> Cc: BOINC Developers Mailing List; BOINC Alpha Mailing List; Steffen Möller
> Subject: Re: [boinc_alpha] [boinc_dev] Boinc alpha and the testing future
> (patchincluded)
>
> The biggest problem I deal with is work fetch.  I'm running into regular WU
> starvation.  GPU work fetch with exclusions and 2 same type GPUs is still
> not fixed.  In addition when using an app_config to limit instances of any
> given CPU app I often find the other CPU app unable to get work with the
> dreaded "not needed" or "project not the highest priority"  messages.  The
> only way to get new work is to suspend work fetch on every project but the
> one starved and then manually poll for it.  A major ongoing frustration,
> but otherwise the CPU cores sit idle.  Can't there AT LEAST be a mechanism
> where when we manually poll for work the client allows it?
>
> Thanks/Ed
>
>
>
>
> On Tue, Feb 26, 2013 at 2:21 PM, Rom Walton <[email protected]> wrote:
>
> > Well since we haven't begun work on a 7.2 release, everything committed
> > since 7.0.28 is technically a bug fix.
> >
> > Even the app_config.xml configuration file (
> > http://boinc.berkeley.edu/trac/wiki/ClientAppConfig) started its life as
> > a power user bug fix.  Ideally, at some point in time, we can figure out
> a
> > sane set of policies that the client/server can use to implement that
> > functionality automatically.
> >
> > I recognize Steffen, Gianfranco, and all the Linux package maintainers
> > have a really hard job.  They are constantly trying to balance BOINC
> > project goals with the Linux distro policies.
> >
> > From my perspective any build in the 7.0b branch past 7.0.28 should be
> > able to be considered stable from a package maintainers perspective.  Do
> I
> > need to change something to better support our Linux package maintainers?
> >
> > ----- Rom
> >
> > -----Original Message-----
> > From: [email protected] [mailto:
> > [email protected]] On Behalf Of Toralf Förster
> > Sent: Tuesday, February 26, 2013 12:36 PM
> > To: [email protected]
> > Subject: Re: [boinc_alpha] Boinc alpha and the testing future
> > (patchincluded)
> >
> > Probably Steffen compares the current BOINC version history + release
> > management with the one of the linux kernel development model (which I
> like
> > too).
> >
> > Said that, he's expecting (and I'd appreciate that too) that 7.0.x is
> just
> > a bug fix version of 7.0.(x-1) rather than a feature-added version,
> right ?
> >
> > --
> > MfG/Sincerely
> > Toralf Förster
> > pgp finger print: 7B1A 07F4 EC82 0F90 D4C2 8936 872A E508 7DB6 9DA3
> > _______________________________________________
> > boinc_alpha mailing list
> > [email protected]
> > http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_alpha
> > 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_alpha mailing list
> [email protected]
> http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_alpha
> 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