[pnfs] Fwd: Class Project Idea: Simulating pNFS at scale

[William A. (Andy) Adamson]

---------- Forwarded message ----------
From: Garth gibson <garth at cs.cmu.edu>
Date: Dec 20, 2006 2:54 PM
Subject: Class Project Idea: Simulating pNFS at scale
To: Andy Adamson <andros at citi.umich.edu>, Rob Ross <rross at mcs.anl.gov>, Greg
Ganger <ganger at ece.cmu.edu>, Benny Halevy <bhalevy at panasas.com>
Cc: Garth gibson <garth at cs.cmu.edu>, Gary Grider <ggrider at lanl.gov>, Lee
Ward <lee at sandia.gov>, David O'Hallaron <droh at cs.cmu.edu>, Peter Honeyman <
honey at citi.umich.edu>, Milo Polte <milop at cs.cmu.edu>

Folks,

Milo and I are planning to run a small grad class at CMU this spring
on HPC storage.  We'll do various readings, for which we'd be pleased
to receive recommendations, but the core idea is a project that I
hope will interest you.

We are proposing a class project to construct a virtual machine-based
simulation environment for executing tests against pNFS [on PVFS] at
large scale -- say 100, then 1000, then 10,000 virtual clients.

We know that the code will not have already debugged at these scales,
so perhaps some of the bugs we will find will not have needed this
scale, but hopefully some of them will.

We know that simulation of 1,000 virtual machines on 10-30 real
machines will not be timing accurate, so we don't know how much
performance debugging we can do, but it seems like a good challenge.

Because resources are scarce, everything is changing, and debugging
someone else's code is hard, I'd like to gather your comments, your
support and your interest in collaborating on this.

This would be a PDSI project, and if all goes well it will live long
after the class ends, so it could become a basis for your testing, if
you are so inclined.

Comments?

garth
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://linux-nfs.org/pipermail/pnfs/attachments/20070117/8b3db897/attachment-0005.htm