Crisis in HPC Discussion - John Pelan, Queen's University Belfast

Newsgroups: uk.org.epsrc.hpc.discussion
From: John Pelan 
Subject: Re: Crisis-in-HPC conclusions
Organization: DAMTP, Queen's University of Belfast
Date: 4 Oct 95 17:06:14 GMT
Message-ID: <1995Oct4.170614.3858@queens-belfast.ac.uk>

In conclusions, P.H.Welch@ukc.ac.uk (phw) wrote:
>CONCLUSIONS: Crisis in HPC Workshop (11/9/95, UCL)
>  o Actions?
>
>    Educate/research/develop/publish/influence.
>
>    Teach high-level models of parallelism, independent of target
>    architecture.  Teach and research *good* models that scale, are
>    efficient and can extract much more of the parallelism in the
>    users' applications.  Priorities are: correctness, efficiency and
>    scalability, portability, reliability and clarity of expression.
>    Maintenance of existing vector/serial codes is not relevant for the
>    long term.

Moreover;

Learn to program reliably and efficiently from the outset. There's no point in porting to more sophisticated architectures with untidy, poorly tested and/or undocumented code. Too much legacy code is like this and it can only serve to impede progress towards HPC 'ports'.
Learn to use standard libraries.
Learn to identify the best granularity for the task in hand so that a suitable architecture can be chosen from;
standalone workstations, workstation clusters,
vector supercomputers, superscalar supercomputers, MPP etc.
Learn to accept that not all tasks are suitable for MPP, say, despite the pressure to work on a 'sexy' machine. One shouldn't be ashamed to use a Pentium cluster (for example) if that's all one needs.
Understand the concept of through-put. A faster (and more expensive) machine *will* be used by more people, which *will* give an effective through-put equivalent to a slower (and most probably cheaper) machine.

--
John P