Crisis in HPC Discussion - Peter Welch, UKC

Newsgroups: uk.org.epsrc.hpc.discussion
From: P.H.Welch@ukc.ac.uk (phw)
Subject: Efficiency again ...
Organization: University of Kent at Canterbury, UK.
Date: Thu, 05 Oct 95 15:40:25 GMT
Message-ID: <176@cypress.ukc.ac.uk>

In article 35 of uk.org.epsrc.hpc.discussion, Lyndon Clarke
(lyndon@epcc.ed.ac.uk) also writes:

|* Efficiency of HPC resources - perhaps I have an unconventional opinion, but
|  I do think that in the case of research council HPC resources the important
|  thing is the research and publications that arise from use of the resources.
|  Really, its a question of how much science (or other, as appropriate) one
|  obtains for the money invested ...

|                       ... and what the money happened to be invested
|  in is largely neither here no there. Are there any data on research
|  publications arising from use of HPC and other central resources in
|  the UK?
|
|* Efficiency of scarce resources - this was also a theme in the
|  conclusions.  The point made was that we must use the resources
|  efficiently in order to increase the response time of resources. -
|  ie so that ou jobs spend less time waiting in queues.  Now, if all
|  of our programs ran, say, four times faster on the T3D then I do
|  think we would all submit four times as much work, and our jobs
|  would spend just as long waiting in queues as they do today.

But this admits the case that efficiency matters. Raising efficiencies from 17% to a (modest?) 68% means that we would be getting four times as much science (or whatever) for our money. Unless, of course, the extra work unleashed on the machine was not really necessary - but there ought to be ways of discouraging that!

As far as the turn-around times for individual jobs remaining the same, this can be discouraged by some `fair' queue management - for example, by preventing one person being on a queue more than once. Then, we can still get through four times the work, with individual waiting times quartered - especially for those who need to use the machine less often (possibly because they need to analyse the results from one run before deciding on the next).

Cheers,

Peter.