Report of the Theory and Computation sub-committee of CASCA, November 2001

===========================================================================

 

Members:
David Clarke 2000-2003
Hugh Couchman 2000-2002
Martin Duncan 2000-2003
Daniel Durand 2000-2002

N.B. Couchman proposes that he stay until the end of 2002 as chair of the committee despite his position as secretary of CASCA if this is acceptable to the Board.

 

Activities since June:

Development of the web resources continues. A summer student hired by Couchman did some work on this as part of her duties and she will continue to work on this during the Fall. A survey was sent out (prompted also by the NSERC reallocation exercise) to try to better understand the issues faced in this arena by the community. A copy of the survey is attached. Several responses were received.

Computing needs, problems and solutions are a rapidly moving target. One theme which is clear is that personnel support for computing is very spotty. This is an area which the web resources will try to target in terms of providing information which caters specifically to the computing needs of astronomers.

Hugh Couchman
24th Nov 2001

 

------------------------------------------------------------------------

 

Dear Colleagues,

During town hall meetings over the last year and in other discussions the issue of computing has frequently been raised as something which needs attention.

The purpose of this message is to understand what are the primary concerns of the community in this regard. The origins of this message are twofold: the Steering Committee for the present NSERC Reallocations exercise and the Theory and Computation Sub-Committee of CASCA. Please take a few minutes to respond to this simple questionnaire with comments which may be as detailed or brief as you wish.

In the following, positions are stated with which you may or may not agree - feel free to respond stating and justifying your views.

 

1. HARDWARE

Let us divide computer hardware into three broad classes: desktop (1-2cpus/<~1GB), midrange (4-16 cpus or large memory >~8GB), high end (>32 cpus).

Q1) Is there a problem obtaining (and a need to obtain) access to hardware in any of these categories? What are the issues that face you: cost, suitability of hardware,...? Please answer with reference (at a minimum) to the following points:

Desktop:
One can buy a very capable high-end PC (>1GHz, 1GBRAM, 200GB disc, high end video etc.) running Linux for C$3,000. Almost all relevant software is available under Linux (a notable exception is a good free Fortran compiler). Perhaps the primary alternative now is a workstation from Sun Microsystems. At this price-point it will be less capable. The more capable Sun systems are substantially more expensive. What justifies the choice of a more expensive solution? Legacy applications, local or personal familiarity with the OS? How important are these factors?

High-end:
CFI has funded many installations throughout Canada. A surprising number of these have members of our community (Astro or GSC17) as primary movers. CFI has mandated that 20% of the cycles on these machines be made generally available. Does this satisfy the community need for systems with up to ~100cpus? If not why not? (There may be arguments for very large systems ~1000 cpus but this is more a question of if we need a *multidisciplinary* national supercomputer centre.)

Midrange:
Several people are constructing ~16 cpu systems out of PCs and fast ethernet for ~C$15-20k - the cost of a decent desktop from a few years ago. Are $s required for this from equipment funds or can(/do) small groups 2 or 3 people get together and fund these? Is there a strong need for the more traditional modest SMP systems (which are more expensive)? If so, for what applications and what is the justification?

Other hardware:
Is the cost of large storage/archiving/backup capabilities significant or difficult to satisfy? (Are the very inexpensive IDE RAID systems that people are building a useful storage solution)? Are there difficulties in handling data from remote facilities (observatories/large computing installations)? Is network connectivity/bandwidth a problem, locally, provincially, nationally, internationally?

2. SUPPORT

Q2) How are your computing activities supported? Support falls into a number of categories; please answer with reference to the following at a minimum:

Systems support:
Do you manage your own desktop, is it done by students or postdocs or by someone specifically employed for the purpose? What are the difficulties encountered at your site in providing these services or having these services provided? Have you explored ways of taking advantage of similarities in hardware/software and networks to ameliorate the burden of the rapidly increasing number of systems? Is developing synergies of this kind feasible, in your opinion, within your group, department, the astro community?

Using parallel systems:
Do you wish to use parallel systems (such as a Beowulf), but feel that you do not have the expertise? What would be required to get you there: pointers to useful references, courses, programmer support, ...?

Q3) Are there other computing issues (difficulties/complaints) which do not fall within any of the above discussions/points which you would like to raise?

Thanks,
Hugh Couchman