[petsc-dev] Number of CPU hours per year spent in PETSc code?

[Mark F. Adams]

On Sep 6, 2011, at 6:58 AM, Matthew Knepley wrote:

> On Tue, Sep 6, 2011 at 5:36 AM, Jed Brown <jedbrown at mcs.anl.gov> wrote:
> Has anyone tried to estimate this?
> 
> The obvious solution is to activate the "call home" feature in PetscFinalize. ;-)
> 
> I thought about this again. The main trouble is that users may not want to share the information,
> and currently there is a culture of secrecy around large runs. I think you have to offer something

This can be anonomized, its done all the time with cell phone data, but you want to make it clear you're doing it and have an opt out feature.  And some users will provide their own 18" air gap opt-out mechanism anyway.

And the fusion guys use several million CPU hours a year.

Stephane: do you have any estimates for this?  CPES gets about O(20) million hours/year as best as I can remember and about 25% of the time is in the solver.  Then there is GTC-P, GTC(irvine), and GTS.

Also, the bone folks that I work with used to be the 2nd largest user at SDSC, but I've not kept up with them.

Mark

> in return. I propose we offer a service like Google Analytics (or www.statcounter.com). That is,
> people autosend performance results and we deliver canned analyses for free, and better ones
> for consulting fees (so that no one will actually do it). Then, after a little while, we make it opt-in
> in configure, and poof we have an idea about a fair fraction of PETSc runs.
> 
>     Matt
> 
> -- 
> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.
> -- Norbert Wiener

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20110906/41b9d449/attachment.html>