[Beowulf] Entry Level HPC?

[Prentice Bisbal]

Tim Kissane wrote:
> I've been reading (lurking?) this list for some time now, because
> of my personal interest in Beowulf systems. The recent discussion,
> "Beowulf SysAdmin Job Description", has prompted me to ask a
> question. What are the entry requirements for a job in HPC?
> 
> I'm 48 years young and looking for work. My resume can be seen
> online at http://timbury.net/tim/ and you will note that I have little
> "official" sysadmin job experience. I started running Linux around
> 1995 with kernel 0.99. I've been an evangelist ever since. I got
> involved with Linux because I wanted a job working with Unix.
> 
> I've since installed, configured, maintained and administrated
> a number of Linux networks as a consultant. I worked over 2 years
> as a release engineer responsible for build management across
> AIX, HP-UX, Solaris, Linux and Windows platforms. Scripting for
> system and admin functions in perl, sh, etc. is no problem.
> 
> My educational background is a Computer Electronics tech school
> way back in '83 and a BS in physics in '92. Some grad studies in
> physics and compsci, but I left school to take a job.

Tim,

Just because you can't afford the hardware, doesn't mean you can't
create a small virtual cluster using VMware or Xen, or a couple of old
boxes. For administration purposes, a cluster doesn't have to be fast,
it just has to be a couple of computers with a queuing system and an MPI
implementation. Heck, even a queuing system is optional (but experience
with SGE or Torque/PBS will definitely be a common job requirement)

Building your own cluster definitely helps. In one telephone interview,
they asked if I had any cluster experience. I said "Yeah, I built one in
my bedroom last night from a couple of old PCs I had lying around and
100 Mb/s hub". That definitely made a difference!

I would read some online books on MPI programming, or Peter Pacheco's
"Parallel Programming with MPI". Even if you aren't expected to do
programming, it helps to familiarize with MPI. It also helps to
understand the concepts of parallelism - what problems are embarassingly
parallel, and what are tightly-coupled. There are plenty of books on
parallel computing that aren't too hard to work through.