[Bioclusters] Apple/Genentech BLAST

[Chris Dwan (CCGB)]

Ivo,

Thanks for your clarifications.  I'm glad we're on the same page.  It
just goes to show the dangers of using "BLAST" as a verb...even among
people who (theoretically) understand what we're doing.

Back to the G4 topic:
---------------------
We've (the University of Minnesota) recently begun populating user
labs with dual processor G4's (2GB RAM, 160GB of disk).  I've been
tasked to integrate these machines into my cluster environment using
some sort of cycle stealing technology.

I plan to make each lab into its own seperate group or flock.  Each
one will have a local file server to handle data issues, as well as to
appease some of the political issues of setting up cycle stealing
between departments.

I've been in contact with the Condor team about an OS-X version, and
they tell me "soon" on a "clipped" version, which means no
checkpointing or job migration.  This is fine for my BLAST rich
environment, since (properly decomposed) these jobs are seconds to
minutes of totally independant CPU time.  I don't need checkpointing,
I need more CPUs.  I don't mind throwing away minutes of CPU time
when someone logs in, if by doing so I can *double* the number of CPUs
at my disposal.

The other contender of which I'm aware is PBS-Pro's "floating
license" system.  I'll  be interested to hear stories from other folks
on this list as we bring up the various queuing environments under
OS-X.  We can't afford Platform's product at the moment (all our
hardware is opportunistic, for budget reasons...so I certainly can't
pay more per-node for a queuing environment) but I'd appreciate any
other suggestions on free cycle stealing queuing environments.

I plan to migrate jobs between the group/flocks using Globus.  

Sadly, things move at a glacial pace in the University environment.
I won't have a chance to get on these machines in a serious way
until late June at the earliest.  At that point, I plan to actually
incorporate them into my processing environment and do my benchmarks
as a side effect of running production jobs on them. 

This is what happens to the priority of hardware evaluation when you
have no control over the next round of purchases.  :)

I'm looking forward to Chris D.'s preview analysis, as well as other
folks experiences with these machines.

Thanks for listening.  

-Chris Dawn