DISTRIBUTION: F R A Hopgood E B Fossey P M Girard C J Pavelin M R Jane A R Mayhook D G House R W Witty D Williams R Taylor K Robinson P G Davey R E Thomas P E Bryant
Over the last year, the Division has put together a policy statement indicating the directions in which we are attempting to go, given the external influences upon us.
It is appropriate now to consider a document which defines strategy over the next five years. It should define a strategy for each of the areas mentioned in the policy statement, namely Service, Support for SERC Research, Contract Work, Education and In-House Research.
At the December meeting of CCC, it was agreed that there should be a complete review of forward planning for computing over the next year for inclusion in its next FYFL. With the CRWP Report only partially implemented, it needs to confirm the currently outstanding parts of that programme suitably modified in the light of past changes in those recommendations (installation of ATLAS 10, for example).
Dr Manning, the Computing Coordinator, would like any document to CCC to touch on issues such as the impact of Single User Systems, the Academic Network, administrative computing, vector processing and the impact of Alvey.
This gives some guidance to the task ahead. I have also received comments from Group Leaders and will attempt to factor these into this paper where appropriate.
The previous attempt at a strategy which resulted in the CRWP Report had a much easier task than we have today. We were in a period of running obsolete equipment, under financed and insufficient resource. All that was required to get a positive response was to define how much should be spent and when. The only issues that really came up were the extent of future demand, the need for vector processing and some discussion of site related activities.
I think the position today is more complicated and any strategy paper needs to make positive statements concerning the following:
It is possible to say that the old values of large scale IBM-style processing have been swamped recently by the euphoria of new ways of doing computing - and yet the need for old-style data processing remains.
SERC will continue to support fundamental research which especially requires great data-reduction and number crunching power. Such facilities will be closely linked to the RAL site particularly in the areas of astronomy, space and high energy physics but also in Science Board activities.
The world is moving to a position where it may, long-term, be IBM compatible. Interfacing and interconnecting to IBM style hardware is becoming common place.
One of the threads of our long-term strategy must be the continuance of a large data processing facility associated with IBM-compatible hardware.
Technical developments will give us plenty of cheap processing power, channel speed and storage. Where are the bottlenecks?
Positive statements need to be made concerning optical storage. Do we wish to pioneer or wait? Shall we base the next five years on MSS and central filestore?
Main memory sizes are going to escalate. The latest Fujitsu is capable of handling 256 Mbytes of main storage. We will find mainframe software making use of this. We need to emphasise the need for continuing to upgrade real memory in line with current practice.
We have a number of incompatible threads to our strategy for distributed computing:
A major change since the CRWP Report has been the attempt to contribute costs of central computing back to Boards. It brings to the front the whole problem of charging.
Bart has put particular emphasis on this in his thoughts and I repeat them here verbatim.
As is well known, Council became concerned a number of years ago about the scale of its provision for computing and about the lack of coordination in its planning. To remedy matters, SERC (or SRC as it was then) set up an advisory body, the Facility Committee for Computing. This has since evolved into the Central Computing Committee, whose remit is the oversight and provision of SERC-wide computing facilities and initiatives.
It is significant that CCC funds have been set aside by Council and indirectly charged to the Boards. A change is proposed (as I understand it) under which CCC funds will be deducted from the Board allocations. This will reopen debate, not in itself a bad thing, on whether the Boards should release so much funds for computing and whether the facilities could be obtained more cheaply in other ways.
I believe that Computing Division's strategy will need to take account of such debates. In particular, cases for provision of equipment will need increasingly to look at cost-effectiveness.
Unfortunately, the CCC is adopting a policy of pushing as much of the funding for computing directly to Boards. We see this very clearly with the decision that the support for AI computing should be undertaken by the Engineering Board. Giving the Nuclear Physics Board freedom to overuse its allocation without extra charge is another example. I believe that these factors, supported by the desire of certain senior SERC officers to attribute computing cost, are beginning to fractionate the SERC support for computing and to bring about less coordination. In fact, I see this as a reversal of the policy which led to the formation of the FCC.
It is true that CCC has never had to oversee all aspects of SERC computing. Developments like STARLINK and the IEC initiatives in DCS, Software Technology, and now IKBS have gone ahead with minimal involvement by CCC. This state of affairs explains why the NP Board questions the rule of CCC in its affairs. As a major consumer of computing, it feels (or so it seems to me) that it has a right to exercise more control over its own destiny.
The climate is ripe for a serious debate about the organisational arrangements for funding SERC computing. The Computing Division needs to formulate its strategy against possible changes:
We are going through a period where SERC is attempting to reduce its direct support for computing mainly as a result of the current Secretary of SERC's initiatives.
This is leading us to a situation where we are getting strong constraints on manpower and a set of projects which have all been under-financed, under-manned and a lack of a clear future. Particular examples are STARLINK, ICF, the PERQ Project. We land up being castigated for not meeting deadlines and yet the fault is with the funding source.
This is a vicious loop which we need to get out of in any strategy document.
Some guidelines for upgrading installed facilities and manning levels must be established. The need for adequate support for activities such as databases needs to be made. The Rutherford philosophy of penalising the Division for its peripheral position on site by charging us directly for functions such as the Receptionist which are clearly Overhead functions needs to be highlighted. The 109 MY of Central Computing is much less when such statements are taken into account.
Another problem in the same area is accommodation. Unless we can recapture the hardware development area currently pirated by Technology Division, the chance of decent hardware developments being pursued in the Division will be low. A strategy statement needs to look seriously at the requirements for accommodation in the future. It is feasible that we shall have inadequate air-conditioned space.
SERC has established a clear need for high powered vector processing facilities for UK researchers. This will mean the purchase of a next generation vector processor for SERC. It is sensible that it is IBM compatible if the power can be obtained that way and it looks British. It should clearly be installed at RAL.
The whole case for a VP, the relationship with DL, joint funding with the Computer Board, and the acceptance of an overall strategy for vector processing needs to be spelled out. How this relates to the need for a high powered interactive processor also needs to be established.
A major new thrust for the Computing Division must be to become the provider of a computer service overall for the site. All site users will want to use centralised services if they provide what they want. We have not gone out of our way to provide it in the past. We need to put it strongly as a major plank in our future strategy. Particular areas to be addressed:
Cliff has proposed a structure for a Strategy document which I reproduce below.
We have a policy, indicating the directions in which we are trying to go. The strategy document should be a five year forward look, saying roughly how we are going about it. It divides up into sections corresponding to policy headings.
Each section has the following sort of information:
There is probably a man-year of work in doing the first one and 3 man-months updating each time (although the FYFL exercise should be much easier).
I have tried to include most of the points raised by individuals in the previous sections. Some additional questions that need answering are:
This paper is really the opening salvo. I suggest we have a brainstorming discussion at least attempting to give some overall guidelines to narrow the alternatives.