[petsc-users] Obtaining bytes per second

[Justin Chang]

For clarification purposes:

1) What is the definition of "performance model" and "cache model"? I see
those two terms used in this thread but want to know the exact difference
if any.

2) Is what's described in Dinesh's paper a "cache model"?
What exactly is the caveat or what are the major assumptions that it makes?

3) Is quantifying the "useful bandwidth sustained for some level of catch"
analogous/related to cache register reuse and/or vectorization (e.g., how
well one can maximize SIMD on the machine if that makes any sense)

Thank you guys for all your help,
Justin

On Thursday, May 7, 2015, Jed Brown <jed at jedbrown.org> wrote:

> Matthew Knepley <knepley at gmail.com> writes:
>
> > On Thu, May 7, 2015 at 1:32 PM, Jed Brown <jed at jedbrown.org> wrote:
> >
> >> Matthew Knepley <knepley at gmail.com> writes:
> >> > You are making assumptions about the algorithm here. Pure streaming
> >> > computations, like VecAXPY do not depend on the cache model.
> >>
> >> I thought we were talking about Krylov solvers, which includes MatMult.
> >>
> >
> > Yes, and in the earlier mail I said to just use the model in Dinesh's
> > paper so he did not have to worry about this one.
>
> That is a cache model (analogous to "perfect cache" in the pTatin
> paper).  So don't claim you have no cache model.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20150508/e2ce1ee0/attachment.html>