[Beowulf] Stroustrup regarding multicore
ajt at rri.sari.ac.uk
Sat Aug 30 07:42:09 PDT 2008
Ed Hill wrote:
> The "roll your own" ethic is arguably THE central theme of beowulf-
> syle computing and it applies to all aspects: hardware, software,
> wetware, ...
> The parts do not exist in isolation. Understanding how things work, how
> they fit together, and how the whole can be improved is a worthwhile
I agree with you - I'm sure I'm not the only one reading this list who
comes from a biological background, but built his own Beowulf: Will the
real Tim Cutts please stand up, and acknowledge that many of us working
in this field are NOT trained programmers. Some of us, including you,
Tim, cut our teeth writing software for our PhD projects, and the task
of trying to explain the scientific problem we were working on to a
'programmer' is more difficult than learning how to program yourself.
In my youth [cough], a 'programmer' was just the person who implemented
the algorithms designed by a systems analyst, and did NOT possess any
special powers. In due course, these people became 'analyst/programmers'
and now we are talking about 'programmers' as if they are analysts...
This list is both useful and interesting to me BECAUSE I built my own
Beowulf cluster, with some advice from Tim at the Sanger (thank you
Tim). In particular I chose Debian, now Ubuntu, after looking at what
major bioinformatics centres like the Sanger use. The controversies
about RH vs. Debian etc. that we've debated here recently are HEALTHY!
So is the debate about specialisation of skills in teams deploying and
running Beowulf clusters. One size does NOT fit all. I've learnt a lot
from reading this list, and it helped me make decisions about what is
practical to attempt on a DIY Beowulf cluster and what is not.
I also believe that building DIY Beowulf clusters is following in the
true spirit of what Donald Becker started. Mine is based on EPCC's (now
dismantled) BOBCAT - "Budget-Optimised Beowulf Cluster using Affordable
This is at the opposite end of the spectrum to some of the 'BIG-Iron'
that many people on this list talk about BUT it is still a Beowulf and
much of what has been discussed here lately is relevant to me!
I worked on an 9216p SIMD array in my last job (over 20 years ago). I
commissioned this machine from a prototype built by a large high energy
physics (HEP) group in the UK. I was exposed to their HEP culture long
enough to know that some people judge a project mainly on the basis of
how BIG the budget is, rather than the quality of the science. In my
experience, some of the very best science has been done without proper
authorisation, and on very small budgets.
Earlier in this thread, Peter mentioned astronomers grinding lenses: It
might be worth mentioning that it was Michael Faraday's unauthorised
experiments done without proper funding at the Royal Institution, where
he was employed to grind lenses at the time, that lead to his discovery
of the laws of electromagnetism and his invention of the electric motor.
My point is that it *is* the science that matters and that building
computers is just a means to an end. Not long ago, many people believed
that RISC processors were the solution to all our problems because a
compiler would optimise low-level RISC code, and there would be no need
to use 'inline' assembly code to achieve maximum performance. Functional
abstraction is essential in programming, as is information hiding. So,
why are we talking about arcane aspects of 'optimising' FORTRAN?
Programmer productivity is just as important as computer performance,
and using tools that allow scientists to express their problems in a
high-level programming language is the top priority. Optimising the
performance of these tools is what I believe 'professional' programmers
are best employed to do. Using them is what scientists do. I've written
image analysis programs myself in assembly code, but I wish I'd had the
high-level languages we now take for granted because I could have done a
lot more science with the time I had available...
Dr. A.J.Travis, University of Aberdeen, Rowett Institute of Nutrition
and Health, Greenburn Road, Bucksburn, Aberdeen AB21 9SB, Scotland, UK
tel +44(0)1224 712751, fax +44(0)1224 716687, http://www.rowett.ac.uk
mailto:ajt at rri.sari.ac.uk, http://bioinformatics.rri.sari.ac.uk/~ajt
More information about the Beowulf