[nylug-announce] NYLUG Tuesday 15 may General Meeting Presents: Eric Moore -on- Computing Clusters

[John Bacall]

                               Eric Moore
                                  -on-
                           Computing Clusters


Computing clusters. Harnessing the power of multiple computers to do
lots of work. This Tuesday, 15 May Dr. Eric Moore will present on the
usability, reliability, computational might of connecting various
hardware to form a single computational entity. A cluster of computers.

You might wonder, doesn't, didn't virtualization render clustering
obsolete? expensive, unreliable by comparison? ``Making one PC look like
many is not the same as making many look like one.  Virtualization is
for when your iron has more horsepower than needed to do one thing, so
you want it doing many things.  Clustering (at least the sort I will
discuss) is for when too much horsepower is not enough (and in my world,
too many flops is never enough, such is life in a field where O(N^4)
algorithms are considered fast, and run times are often measured in
weeks),''says Eric.

Does Linux lead in clustering? Doesn't *BSD rockit harder? ``Yes, no.''

Should you get Slackware, or Ubuntu, or that little book that tells me
how to roll your own Linux?

``Depends on what you're trying to do.'' Eric will share his experience
and astutely guide your options.

Clustering is basically the art of harnessing together many computers to
do computations that can't be done by one (or in many cases, to
conveniently do a bunch of computations that can be done on one all at
the same time).  (There are also high-availability clusters, which will
be discussed only peripherally). Clusters are found in all sorts of
places, from Google, which uses them to build its search indexes, to the
giant render farms that give us animated movies, to the beowulf clusters
of scientific research (and their Ghettowulf cousins which Eric has
mostly built).  If it takes more than a couple hours for your program to
run, you may want a cluster for that.

The talk will cover a number of questions regarding:

* Can I use a cluster for my problem?
* Should I?
* What sort of cluster?
* What tools are there for my kind of problem?
* What are the trade-offs I need to look at, and what works in what
  situations?
* What are my options?
* Who uses these things, and for what?

Clustering is a really big field, and covers a lot of ground, and
there's a lot of really good stuff for doing it under Linux, come hear
about a whole bunch of these things.

If you need a cluster, think you need a cluster (or think you don't need
one and want to make sure), or just have a bunch of PC's you want to
turn into a cluster just because, you just might learn something.

It never hurts to come to a talk about an area where Linux and free
software are the undisputed masters of the field.

RSVP your place at the marvelous Google lair for this Tuesday, 15 May
NYLUG meeting. Many thanks to Google for their kind hospitality in
hosting our group.

Further Information:

RGB's beowulf book:
http://www.phy.duke.edu/~rgb/Beowulf/beowulf_book.php

The beowulf website:
http://www.beowulf.org/

OSCAR:
http://oscar.openclustergroup.org/

Open Mosix:
http://openmosix.sourceforge.net/

SGE:
http://gridengine.sunsource.net/

Condor:
http://www.cs.wisc.edu/condor/


About Eric Moore:

Eric's a computational chemist, who has had to set up and maintain
several clusters in his career, and has learned most of the ways not to
do it, and some things to do.  Right now he teaches for a living, but
would rather be crunching numbers.  Giving a talk about crunching
numbers is almost as good.