The appearance in the market place of cheap high powered single user system with good interactive capabilities via high precision displays heralds, completely new way for the average user to achieve the major part of his computing requirements.
Within the next few years, many such systems will be available from different manufacturers. Initially, it is likely that they will be available from small companies and the level of software support will be negligible, Consequently, there is a likelihood of many different systems being purchased in the SERC environment and a great deal of duplication of basic software development.
We see a need for a coordinated development plan to ensure that the UK makes the best use it can of its limited manpower.
It is essential that UK industry gets established quickly in this market and that purchases of systems are made from a UK manufacturer. As there is no UK product, we urge that the ICL initiative to manufacture and market Three Rivers PERQ computers is strongly supported by bulk SERC purchasing. SERC should work with ICL and DOI in establishing ICL as a major force in this market place.
There have been enormous advances in the scale and ways of using computers over the last 20 years. In particular, there have been significant changes in usage which have come about because hardware and software advances have allowed quite revolutionary ways of tackling the problems current at the time. In the early days, computers ran a single job at a time. The first major break through was to allow multi-programming with many jobs sharing the resources of a single large system. The main reasons for this approach were the need to utilise expensive resources to the maximum (large computers were the most cost-effective as long as they did not stand idle) and advances in software which resulted in the concept of a supervisory program or operating system to schedule the resources of the system.
In the 1970s, the second major advance in computing was the development of interactive systems which allowed many users to access a computer so that the computer appeared to be their own private system. Unlike multi-programming where jobs were placed in a queue to be run later, the user was able to develop programs at a terminal and to interact with a program that was executing. With the ability to input and output graphical information as well as text, large scale computer-aided design systems developed. Interactive time sharing systems appeared partly because new software concepts such as paging and virtual memory allowed efficient operating systems to be developed but also because technological advances such as large memories made it economic. Even in the 1970s, it was, in general, too expensive to allow a single user the sole access to a computer system except in very special circumstances. Also, an advantage of many users accessing a single system is that program and data can be shared which is particularly advantageous when many users are collaborating or involved in a single project.
In the last few years, a number of major hardware and software advances have made it possible for a third major change in computer usage. A single user system is now an economically viable way of providing computer power in the future. Advances in networking and inter communication make it possible for these single user systems to be interconnected to retain all the advantages of the large main frame in terms of program and data sharing. This concept of Distributed Interactive Computing will have a profound effect on the way computing advances in the 1980s.
SERC, with its earlier initiatives in the Interactive Computing Facility, The Distributed Computing Systems coordinated programme of research, and the recommendations of the Roberts Panel, is in an ideal position to make a major impact in the way computing develops in the UK in the next decade. It is essential that a coordinated programme of development within the SERC environment is achieved to maximise the benefits of this new approach.
In this paper, we recommend that SERC should launch a specific programme involving central purchasing, standardisation, software development and collaboration with UK industry and DOI.
About six years ago, the Xerox Palo Alto Research Centre (PARC) developed a single user system called the Xerox PARC Alto. These systems were linked together to form a powerful research facility which demonstrated that a single user system located in a researcher's office provided a type of computing that gave significant benefits over modern time sharing systems. The terminal could take the place of many of the functions in a conventional office, including the desk and mail service, as well as allowing a mode of working which was richer and more flexible than that available through conventional time sharing systems. It also allowed increased scientific productivity. The Xerox development was a research project that was uneconomical at the time but pointed the way forward.
Within SERC, the Interactive Computing Facility made the decision that, due to the decreasing cost of memory and cpu power, the most cost effective way of providing interactive computing was not via large central mainframe systems but via smaller multi-user systems situated in university departments. The Rosenbrock Report, which recommended the formation of ICF, anticipated that within the next decade, the cost of computing hardware would decrease significantly so that in the long term single user systems would become cost effective. The ICF has realised the need to communicate and provide access to central specialised facilities. Consequently, the multi user systems in university departments are connected via SERCnet to allow university researchers to access any of the facilities on the network.
ICF has purchased a number of low-powered single user systems such as the IBM 5100 for evaluation. Although providing a new way of doing computing, these systems lacked power and have been of limited value.
The Distributed Computing Systems programme has been concentrating on the problems associated with linking a number of processors together to form a coherent computer system. Until recently, this research has been based on cheap low powered systems having only some of the characteristics of the Xerox PARC Alto.
The main components of a single user system are:
Three years ago, a system with these characteristics could not be purchased at a cost less than 70 Kpounds. There have been a number of significant advances in the intervening period which now make such systems economically viable:
These technological advances mean that single user systems can be built and marketed within the cost given above. Also, many of the technological advances are still in an early stage so that decreases in cost of such systems or increases in functionality at the same cost are likely in the near future.
In May 1979, Three Rivers Computer Corporation, situated in Pittsburgh, announced a system, the PERQ personal computer system with the major characteristics outlined in Section 3. Full details are given in Appendix 4. It was clear that a commercially available system at reasonable price would have a significant effect on the UK research scene. Both ICF and the DCS programme showed immediate interest and Rutherford Laboratory were the first customer to place an order for a system. This was done via the British Embassy in Washington in June 1979.
Three Rivers is a small company of about 30 employees which has, in the past, specialised in the production of high quality graphical equipment. It marketed an electrostatic refresh display in the mid-1970s which was one of the fastest displays on the market at that time. More recently it developed a high quality colour raster system which has also been well received by the market place.
Brian Rosen, the developer of the PERQ, was previously at Xerox PARC. PERQ was officially launched at SIGGRAPH in Chicago (August 1979).The prototype system was on show but with very little software. The company had hoped to ship systems before the end of 1979 but were much too optimistic . They had problems with their suppliers of PCB boards and lack of finance slowed down developments.
A major change in the Company came with the arrival of Prof Fredkin from MIT initially as consultant and later as Chief Executive. Prof Fredkin was a major force in III, the makers of the Rutherford FR80 microfilm recorder. He raised $5M for Three Rivers from major banking houses for 23% of the company. Near the end of 1980, the company moved from a small factory to a four storey building of about 20,000 sq ft.
It was clear that once it became known that such systems were on the market, SERC would get many grant applications requesting this system and this would involve the SERC in the purchase of a considerable amount of USA equipment. Consequently, Rutherford approached ICL, the major UK computer company with a proposal that ICL should negotiate an agreement with Three Rivers to market and, if possible, manufacture PERQ systems in the UK. ICL responded positively. Consequently, papers were submitted both to ICFC and Computing and Communications Sub-Committee in early 1980 with a number of recommendations that were accepted by both Committees:
Both Committees agreed that these recommendations would provide support and encouragement for a timely initiative by British industry and provide a practical stimulus to UK research.
ICL started negotiating with Three Rivers in February 1980. Several executives visited the company and engineers from their Utica plant in the USA did a thorough evaluation of its reliability and cost. Negotiations continued throughout 1980 but were hampered by the lack of a PERQ system in the UK to show ICL management (a full calendar of major events is given in Appendix 1).
The first PERQ systems arrived in the UK at the end of 1980 and there are now a total of 5 systems in the UK - 3 at the Rutherford Laboratory (one owned by ICL) and two in industry (Logica VTS and Graphical Software Ltd). Progress since the delivery has been quite fast.
Prof Fredkin visited the UK in March 1981 and ICL now has a draft agreement with Three Rivers to market PERQs exclusively in the UK, South Africa and Australia and has an option to manufacture PERQs in the future. This should be finalised before the middle of the year. It is timely to consider what attitude SERC should take to a major new thrust in the evolution of computing and, in particular, what area's of SERC's activities require these facilities.
The importance of the ICL initiative is that the UK will be in a position to become involved in this area of development. Experience of such systems and how they can be used needs to be looked at in depth. Discussions have also taken place with DOI who welcome the ICL initiative and see SERC's role as being of importance particularly in the area of software development.
As has been mentioned above, a number of bodies within SERC have already shown interest in the PERQ. These include the following.
PERQ is seen as an excellent single user system for interactive working. In some areas, computer aided design can be accomplished completely within the terminal - in other areas it will be used mainly for the pre end post processing stages with the network connection providing access to the large mainframe systems for detailed analysis.
The Artificial Intelligence community is particularly interested in the PERQ for two major reasons. A great deal of their research requires rich interaction in that they are particularly interested in the psychological aspects of the man-machine interface. Also, the AI community tend to use highly specialised languages such as LISP and POP-2 which can be made much more efficient if specific list processing and stack manipulation orders are added to the order code of the system. This can be done in the case of PERQ by extending the order code of the machine.
The DCS programme is particularly interested in the properties of coupled processors and the problems associated with distributing processor power and operating system across a linked set of systems. A particular area of research funded by DCS is in Office Automation and there is already outstanding a grant to QMC for two PERQ systems for this work.
Information Engineering Committee have agreed to a coordination of grants in the area of Software Technology as part of the Roberts Panel initiative. Coordination will be provided from the Rutherford Laboratory and the Committee has agreed that the PERQ would be an ideal vehicle for this research.
The power of the PERQ is such that it could provide a vehicle for program development and testing in place of the IBM and Multi User Mini front end service. A major use of the PERQ could be in this area of software development. Once tools are available, it would be of significant use in those areas of the Council's activities involving large scale software projects. These occur in all Board's programmes. With the falling cost of hardware, software costs will tend to dominate the 1980s. A significant proportion of SRC's manpower and funds are used in software development. It is becoming increasingly important that programmer productivity is maximised.
SERC has a number of major projects such as SNS where high speed computation facilities linked with good display facilities and interfaced to instruments are required at the data collection point of the system. The PERQ appears to satisfy many of the requirements in this area and its applicability here needs to be assessed.
Most of the office automation system on the market today do not have the processing power or display quality to be an adequate replacement for the office desk. The quality of the PERQ display, the large local storage and the power of the vehicle make it an ideal system for office automation applications.
High powered processors with a good man-machine interface will have an inherent flexibility and interest which will make them attractive vehicles for a whole range of research activities. This area will probably be the largest and most uncoordinated. For example, in the recent Computing and Communications Sub-Committee grant round, there were requests for 19 sing user systems of 10 different types.
The PERQ is significantly better than any other single user system on the market at the moment with a price below $40K. This is partly due to Brian Rosen's earlier involvement in the work at Xerox but also due to the background of the Three Rivers company. However, as none of the individual entities that make up a PERQ are difficult to purchase, it is clear that competitors to the PERQ will appear.
The Rutherford Laboratory has kept a close eye on the market over the last year and there have appeared only two systems that are similar to the PERQ but with major deficiencies. These are the Apollo DOMAIN computer and the Symbolics Corporation LISP machine. No British company has a product remotely equivalent to the PERQ. The Apollo DOMAIN is similar to the PERQ in power but has failed to integrate the display closely with the system so that the ability to make swift changes to the display are not there. The Symbolics LISP machine has similar functionality to the PERQ and probably more software but is over twice the price.
At the moment, PERQ has a clear lead over its competitors but they will not sit still. It is clear that rivals to the PERQ will appear and this will fragment the market unless some consistent purchasing policy is established by SERC.
We firmly believe that there is an urgent need for a policy on the purchase of single user computer systems within the SERC. The cost of these systems is sufficiently low that they fall within the signing powers of a large number of individuals. Unless action is taken they will be purchased through establishments and through grants in large numbers and without prior approval. The computing power they provide will be very large but without some control, coordination and standardisation there will be a large amount of wasted effort on unnecessary duplication of effort on software. A large fraction of the time of a major part of the student and research assistant effort in many areas could be spent on inefficient provision of software. It was as a direct of such fears that the ICF was set up. The situation with the single user systems will be an order of magnitude worse just due to the low cost and high performance of these devices but at the same time a small amount of effort provided in a controlled way should enable great benefit to be gained.
It is very important that a strategy be devised and accepted that provides control and guidance without being totally rigid and bureaucratic. A possible strategy is outlined below.
DOI is keen that the ICL initiative is successful and sees great. benefit in ICL and SERC having a collaborative agreement to develop software for the PERQ. They have already indicated their willingness to contribute to such a project. We would recommend that all purchases of PERQ systems in the SERC environment from ICL should be through a central organisation at the Rutherford Laboratory.
By ordering systems in advance, it should be possible to lessen the delay between grants being awarded and equipment being available. By purchasing centrally, significant discounts should be available to SERC that would not be so otherwise (current negotiations suggest that there will be discounts of the order of 25%).
The software available on the PERQ at. the moment is quite rudimentary consisting of a good PASCAL compiler and Editor plus a basic operating system. We recommend that a project team be set up to coordinate the software development on the PERQ in conjunction with ICL. A proposed software plan is given in Appendix 2.
Coordination is required for two classes of users - the application programmers and the system developers. The first needs user support in much the same way as ICF operates now. It is likely that Special Interests Groups will be set. up in specific areas. The system developers need to be coordinated in much the same way as the DCS programme works - mailshots, workshops etc. We anticipate a build-up of software associated with these systems over a number of years. It will be important that existing standards and interfaces will be adhered to by new hardware and software developments.
Once Local Area and Wide Area Networks become widely available, systems should be attached to Local Area Networks so that centralised support can be provided.
Failure to provide a coordinated software initiative will mean that a great deal of research effort will be wasted in providing similar software facilities in a number of different groups. This has already happened with stand-alone systems supplied on grants. The low level of support provided on the PERQ initially, the ability to change the microcode, and the low level of funding necessary to purchase a device will make the situation an order of magnitude worse with these single user systems.
The main impetus so far has been via the DCS programme and ICF, both funded by the Engineering Board. In the future, this is the wrong place for this work to be coordinated. The impact of single user systems will effect the activities of all Boards and it is important that a Council-wide coordination is achieved.
We recommend that the Central Computing Committee be responsible for coordinating the area of single user systems. It may be sensible for this body also to coordinate the work in Local Area Networks which is intimately connected with the Single User Systems.
The Committee must be given sufficient powers to ensure that individual Boards adopt a common policy. All Boards are represented on the Central Computing Committee so this should not be a problem.
At some stage, it will be necessary to define the type of system included in this coordinated plan. We do not anticipate that it will include any multi user systems nor will it include systems predominantly concerned with on-line applications. In broad terms, the system must be one that is designed for high quality interactive use. For this reason, we would not include low cost microprocessor based systems such as APPLEs and PETs in the programme. A clearer definition must be provided at some stage.
There is an urgent need for the project to be established as early as possible - We have a unique opportunity to provide coordination and a unified plan before the situation is past saving. This is a rare situation in SERC's computing policy. In the past, attempts to standardise have tended to be retroactive and, consequently, difficult to achieve.
We do not envisage the standard Board grant applicant being funded centrally but through the grant line of the relevant Committee. However, we do envisage that the grant funds would be transferred to RAL for central purchase of systems and support. SERC should retain ownership of the systems and provide them for the duration of a grant to a particular researcher - The system should then be returned to RAL and reassigned to another grant holder - Boards would allocate systems up to the number they had purchased.
However, there will be a need to provide some manpower and funds for centralised software development and manpower to manage the facility. A suggested expenditure profile for the FYFL is:
| 1981/2 | 1982/3 | 1983/4 | 1984/5 | 1985/6 | |
|---|---|---|---|---|---|
| MY | 5 | 5 | 5 | 5 | 5 |
| Staff Costs (£K) | 100 | 100 | 100 | 100 | 100 |
| Capital | 200 | 50 | 50 | 20 | 20 |
| Recurrent | 100 | 100 | 50 | 50 | 50 |
| TOTAL | 400 | 250 | 200 | 170 | 170 |
In 1981/82, it is feasible that a significant part of the funds required could be obtained from existing programmes who have already shown interest. Until final allocations are available, the extent to which this can be done is still unknown.
The Capital Funding in the early years will provide about 13 systems for software developments intra-murally and via EMR contracts. Recurrent in the early years includes funds for the contracts while it is mainly maintenance and support. in later years. We anticipate that Boards would pay for their fair share of any maintenance costs even if they were provided centrally. The amount of effort defined here provides the management function and some software effort. We would anticipate additional effort being provided by the various Board programmes who have already earmarked funds for work in this area and would be prepared to contribute to this overall initiative.
This paper has set out a plan for an overall strategy for Single User Systems and Distributed Interactive Computing. We invite this Committee to endorse the following recommendations:
The software provided so far by Three Rivers consists of a rudimentary filestore, a full screen editor, PASCAL compiler and loader - Three Rivers have plans for extending the software in various directions. It is clear that, with their limited effort, significant areas of software required by SERC will not be available in the near future.
This paper indicates software developments that will make the PERQ more suitable in the SERC environment and, hopefully, are complementary to the efforts of Three Rivers. A full software development plan is expected from Three Rivers this summer and the SERC plan will be modified in the light of the statements provided by Three Rivers.
Discussions with DOI have indicated that they are particularly interested in SERC's plans and how they make the PERQ more viable as an ICL product. ICL also has effort in the software in the software area and it is anticipated that this paper will form the basis of a combined programme of work by SERC and ICL.
A major difference between using the PERQ and a conventional main frame for interactive work is that it is a single user system. Consequently, the environment does not have to share resources between different users although it will still require to share resources between different tasks initiated by the single user. In the long term, we anticipate that programming and the software environment on systems of this type will be significantly different from the current position. In the programming area, languages are likely to be less procedural with the emphasis placed on the functional requirements of a problem rather than its solution methodology. In specific application areas, there will be a move towards predefined modules and application-specific workbenches so that large application suites can be generated from the problem description semi-automatically.
At the software environment level, we will move away from computer oriented systems and more attention will be placed on the human engineering of the user interface.
Any software plan must aim to provide a maximum benefit to users initially and a graceful transition towards the requirements of the future.
In the 1980s it is not enough just to provide an operating system on a new computer. Users such as applications developers on the ICF CAD community expect a much richer set of facilities including a variety of compilers, software tools, graphics packages, subroutine libraries etc. All of these tools must be integrated to form a harmonious environment in which the user can develop his applications program and/or solve his problem.
The major parts of that user environment are:
In the short term, our aim will be to concentrate on (2) and (3) in way that will allow us to capitalise to the maximum on existing software in the areas (4) to (6).
The operating system choices are:
The Three Rivers P.O.S development has only just begun and there is little existing software. The plans as we know them so far are not very impressive. They have only a small, relatively inexperienced software team and it is unlikely that the P.O.S system will be comparable in facilities to the existing systems in use (PRIMOS, OS4000, etc).
It is feasible that a joint software development plan could be agreed with Three Rivers. However, the geographical distance makes it difficult to achieve. In general, developments of this type in the past have been where the manufacturer had a strong initial product and the Laboratory mainly attacked specific areas which were relevant in the SERC or UK environment.
We do not see this approach as being very viable in the short term. For example, Three Rivers do not have the effort to provide a FORTRAN77 compiler and the software development needed would be considerable.
In the long tern, either by evolution or by a new effort, this will be the correct approach. The single user systems are sufficiently different that this is the only way we can exploit all the advanced features of the PERQ (bit map display, high quality inter-action, networks of single user systems connected by high speed local area networks etc). Such a fully interactive, distributed operating system is a state-of-the-art research and development project which will take several years to complete and would benefit by some widespread use of these single user systems to estimate their long tern potential and ascertain their peculiar requirements.
We see it as inevitable that an operating system of this type must be developed for widespread use in the long term. However, it would not be sensible to base our short term requirements (1980-85) for a sound basis for PERQ software development on such a product. There is a need for an operating system now which would be developed in parallel with the long term solution.
It would seem that the only sensible short term solution is to move an existing operating system to the PERQ. This has two major advantages. At least some users will be familiar with it. The second and major advantage is that it will be reliable and will bring with it additional higher-level software.
The major problem is that moving existing systems is not easy in general in that systems have not been designed to be portable. The major exception to this rule is UNIX which is accepted as being a good interactive operating system and runs on several manufacturer's hardware with different characteristics.
A second advantage of UNIX is that it is the standard environment for Computer Science research and is a de facto standard in the DCS programme. UNIX has also been recommended as the common software base operating system for the Roberts Software Technology initiative.
Finally, the RAL Computing Division has experience of UNIX in that it runs on the PDP11 system installed in the Atlas Centre and is also available on the 3032. The UNIX software tools environment is also supported by the PRIME system.
It is recommended that a UNIX system is developed for the PERQ as soon as possible. The aim would be to produce an initial system quickly even if that means it is inefficient. This would be followed by a more polished version at a later date.
UNIX has a secondary advantage in that the system has with it some of the best software development tools available already. These range from editors and document formatters to FORTRAN pre-processors, command decoders and program checkers. These to a large extent are designed to run in a batch mode. There will be a need to develop genuinely interactive tools such as syntax-checking editors, command decoders allowing graphics tokens etc. There is a considerable amount of medium term work if the interactive potential of the PERQ is to be achieved and the increased user productivity realised.
On the language front, there are many compilers in existence for UNIX. Most of the system itself is defined in terms of the high level language C and the major problem in transferring UNIX to a new machine is the development of appropriate code generators for the C compiler. Once this is done, other languages become available - in particular, a FORTRAN 77 compiler is available written in C.
We have divided the software programme into three phases:
We have not included in this list requirements such as hardware diagnostics which we assume will be supplied by ICL. We would anticipate that the breakdown of manpower needed over the next year or so for the tasks above would be:
SERC 5my Development Contracts 3my ICL 4my
The main requirements are:
To a large extent, the long term requirement is open ended. A major requirement is the development of a fully interactive, distributed operating system to replace the UNIX single user environment. The other main area long term will be to improve the user interface capabilities. Specific examples of work in this area are voice input/output, recognition of characters drawn on the tablet etc.
This paper outlines future developments past the existing PERQ system. The current PERQ has been designed as an effective single user system in stand-alone mode and also as a member of a Local Area Network where servers provide resources unavailable locally.
In some areas, the current PERQ system has more capability in particular directions than is required. Consequently, we anticipate variants of the standard system which are configured for a particular environment at lowest cost. The current PERQ does have deficiencies in the hardware and, in parallel, we anticipate improvements aimed at providing greater functionality at the same price or a little more.
As the PERQ market develops in the 1980s, it is envisaged that the cost of software associated with a system may be the dominant part of the cost. Consequently, new PERQ hardware developments are likely to be constrained to be compatible with existing or planned software.
We will divide the discussion into three parts:
The major changes that we see are feasible in this direction are:
We feel that any decrease in the performance of the processor or display will seriously degrade the attractiveness of the PERQ and do not advocate a lower-performance device.
The PERQ currently has 20 bit addressing, segmentation, and an intimate connection between the memory and display. Peripherals are connected via a Z80 peripheral controller.
The major changes we anticipate as being required are:
The current display is a high resolution 768 by 1024 black and white display. Extensions we envisage are:
Two enhancements of limited value but useful in some applications would be:
The major area in which the PERQ can be extended is to increase the flexibility of the user interface. Some examples are:
