Paper to the Interactive Computing Facilities Committee
The last general report on ICF was the Review of Phase I (EB 78-62, 14.12.78) accepted by the Engineering Board and later by Council in March 1979. It reported progress in establishing ICF since its inception with the Rosenbrock Report (November 1975) up to completion of Phase 1. Phase 1 had been based on central interactive machines at RL, two DEC10 KIs at UMIST and ERCC and seven upgraded interactive minis at University sites. An assessment exercise to choose standard MUMs for new purchases had been completed, an authorisation/accounting system set up, a newsletter established, and 7 Special Interest Groups enrolled to advise on applications software. The paper also made proposals for a Phase 2, in which it was planned to install 15 new standard MUMs and increase the total number of users served to an estimated 1700 by 1980. It was estimated that by 1982 we would see the solution of the technical problems of interlinking all ICF's own machines to each other and to central batch computers, and that support of 9 Special Interest Groups would continue.
The purpose of this paper is:
Twenty-three computers are wholly owned by ICF, including 3 to be installed by the end of 1980, and one to be disposed of in mid-1981. The sites are listed in Table 1: they can be grouped as follows:-
DEC10 2 GEC 4000 series 12 PRIME 750 and 550 9
The 36-bit DEC10 provision is seen as increasingly specialised towards users requiring the widest range of special languages, particularly for AI and some digital design and CAAD research. After 1981 only one KL-10 will be supported by ICF.
The GEC 4000 series machine running the OS operating system has been the normal multi-user mini provided, except where demand was particularly severe, when PRIME 550 (7) or PRIME 750 (4) were installed. All the PRIME systems run the identical PRIMOS operating system.
An additional nine node computers are linked to ICF and run standard operating systems maintained by ICF staff:-
GEC 4000 series 7 PRIME 400 2
ICF has a part share in six more upgraded MUM installations which are also listed in Table 1.
On all the machines partly or wholly owned by ICF, the host site is given a share of the computing resources for its own use in return for a contribution toward running costs. At present this share ranges from 20% (on 16 new MUMs) to 50% (some upgraded MUMs) and is left entirely at the University's disposal. SRC's Contribution toward running costs falls steeply between the 3rd and 5th year after installation to an average level of about £14K p.a. at each site.
Notes on the status of connections between the above machines are given at the foot of Table 1.
The long term aim is to establish a fully linked, compatible network based on the international accepted standard protocols, X25 etc. Connections are required for:
Three IBM 5100 APL series and two Tektronix 4051 machines were purchased in support of SRC Grants following an assessment of portability, and have been giving useful service, particularly the IBM 5100 for support of a research engineer requiring a high-quality APL service. The use of APL seems likely to increase slowly, but in some disciplines interest is high and could lead to rapid developments. It is thought that widespread use of SUMs by ICF should await proving of the Three Rivers PERQ, Apollo, or similar machines offering considerable processing power, very high quality graphics and at least 1 Mbyte of main storage for a price around £12K; and development of a cheap networking connection for it.
The number of ICF users with current registration averaged over 1700 during the period July 79 - June 80 (see Figure 3). Of these, about 1200 have recently been active in a given month. About 70% of these users comprise SRC authorised grants, pump-priming allocations and SIG/systems activities.
Phase 2 has seen a steadily increasing use of SRC central batch facilities via ICF MUMs, and the recent provision (see Table 1) of ICF operating systems on enhanced batch workstations at UCL and Manchester. This trend is highly satisfactory and will accelerate with use of the recently installed DECnet/SRC X25 gateway, and with completion of PRIME networking expected early in 1981.
The AP120B array processor (installed on a PRIME 750 at RAL) has proven useful in a relatively small number of applications - such as 2-D electromagnetic field problems - where it has been worthwhile investing considerable effort in overcoming initial difficulties with the manufacturer's software. This situation is improving as more systems are sold for general-purpose use. Currently eight groups are supported on the AP120B.
The earlier SIG's to present reports to ICFC did so in 1978/79. These were:
Application software developed for these SIGs by ICF staff at RAL and Edinburgh University has entered quite widespread use especially in AI (special languages), Electromagnetics (packages for field and eddy-current calculations in 2 and 3D) and Finite Elements (packages for stress analysis and library of subroutines and problem examples). Initial work for the Circuit Design SIGs has comprised mounting a number of packages for analysis and simulation. Courses run both at RAL and at remote sites on all this software are often over-subscribed.
SIGs reporting later included:
Work in all these areas has started, but no software has yet been released to users.
A full account of application software developed so far is given in Section 6 below.
Site | Machine | Comment | Network Status |
---|---|---|---|
Computers wholly owned by SRC and installed by ICF: | |||
Edinburgh | DEC 10KL | C | |
UMIST | PRIME 750 | (DEC 10KI) | B (C) |
RAL | PRIME 750 | × 2, Prime 400 | B |
RAL | GEC 4085 | A | |
Bristol | GEC 4085 | A | |
Cambridge | GEC 4070 | A | |
Glasgow | GEC 4070 | A | |
Newcastle | GEC 4070 | A | |
Cranfield | GEC 4070 | A | |
Cardiff | GEC 4070 | A | |
Bradford | GEC 4085 | A | |
Birmingham | GEC 4080 | A | |
Sheffield | GEC 4085 | (to be installed) | A |
Herriot Watt | GEC 4085 | (to be installed)* | A |
QMC | GEC 4085 | (to be installed)* | A |
Sussex | PRIME 550 | B | |
City | PRIME 550 | B | |
Surrey | PRIME 550 | B | |
UCL | PRIME 550 | B | |
Warwick | PRIME 550 | B | |
Computers linked to ICF and running ICF operating systems: | |||
Appleton (Slough) | GEC 4070 | A | |
UCL Enhanced Workstation | GEC 4082 | A | |
Manchester Enhanced Workstation | GEC 4082 | A | |
RL Data Editing System | GEC 4080 | A | |
ROE | GEC 4082 | A | |
NERC Keyworth | GEC 4070 | A | |
NERC Swindon | GEC 4070 | A | |
Upgraded systems partly owned by ICF: | |||
Nottingham | PRIME 400 | B | |
East Anglia | PRIME 400 | B | |
Sheffield | Interdata 8/32 | (ICF to withdraw from 1983) | - |
Swansea | PDP 11/45 | - | |
Oxford | VAX 11/780 | D | |
Southampton | PDP 11/45 | - | |
Leeds | VAX 11/780 | D | |
MGU | PDP 11/45 | E | |
Notes on networking status:
* If 1981/82 ICF allocations permit. |
The GEC 4000 machines have between 0.5 and 1 Atlas power, although there is a factor of about 1.5 between the top and bottom of the range. The PRIME 550/400 processing power is 2 Atlas and the PRIME 750 is 4 Atlas power. Both ranges have highly reliable hardware and software.
Both the computers' performances are very sensitive to the size of store, numbers of users and the type of use. The expectations of users are constantly rising and the current estimates of a well balanced computer are:
GEC 4000 512Kbytes 6-10 users PRIME 400/550 1Mbyte 10-15 users PRIME 750 1.5Mbyte 15-20 users
It should be remembered that the faster machines also tend to run larger and more demanding jobs.
The principal hardware deficiency of the GEC 4000 is the maximum 16Kbyte segment size which causes FORTRAN programs with large arrays to be inefficient in many cases. The GEC operating system is sophisticated and provides a good user environment. Extensive enhancements have been made to the operating system which have:
The operating system is efficient and economical requiring only 80 Kbytes of locked down memory. Particularly successful has been the central support of the systems using the SRC network to downline load new versions of the operating systems.
The PRIME operating system has a simple and effective structure with many of the sub-systems (for example the batch system) running as pseudo users (phantoms) rather than being integral with the system. While this insulates the system from the malfunctioning sub-systems, it can be inefficient.
The SRC network offers 3 protocols, ITP for interactive use, FTP for file transfer, and HASP for job submission to the IBM computers. Major extensions have been made to the operating system in the areas of job control languages, accounting, performance measurement and networking. The GEC computers offer all the above protocols. The GEC 4000 series as well as being used for multi-user minis is also used for SRC network nodes and enhanced workstations. Consequently, it was the first system to be fully networked.
Some of the PRIME computers are currently linked via PRIMEnet with a local HASP link providing a direct connection from the RL PRIME to the batch facilities. There is also an asynchronous FTP to provide connections with small local computers and the GEC systems. The PRIMEs will offer SRC network connections with ITP and FTP by the end of 1980. It is not intended to offer HASP over the SRC network from PRIMEs. This will be achieved by file-transferring IBM jobs to a GEC computer and submitting the job from that computer. Techniques are being developed to allow output to be routed back to the PRIME, although this will not be automatic.
The GEC computers are connected to the SRC network using binary synchronous link level protocol. The PRIME computers will use HDLC. The speed of connections vary from 2.4 to 9.6K bits/second. With the networking of the DEC10s and PRIMEs there will be less need for direct asynchronous links and consequently more bandwidth can be provided for computer to computer links.
The Asynchronous FTP is also implemented on LSI11 under RT11, PDP11 under RSX11M, M6800, Intel 8080 and Alpha 16. This has been a low level activity and requires some effort to complete, although, in its current state, it has solved many connection problems: it is the only means by which real-time activities can be connected to ICF machines.
The principal problem with user support has been the low level of manpower available for this activity. The GEC support is given on two levels. First-level support is provided by the Site managers and a second level by two central staff. With 16 machines the amount of time which central staff can given to any one problem is limited, especially since they are also responsible for delivering software to sites via the network. In general, urgent queries are posted via the network to Rutherford Laboratory and at least looked at every day, whilst once a week each site's support file is listed and examined for any necessary action.
The PRIME support has been more difficult due to the lack of network access to the machines from Rutherford and also by the loss of the one full-time support person. Two upgraded sites are substantially self-supporting while the UMIST site is a special case: the intention is for systems support of remote sites eventually to become the responsibility of UMIST. With the networking of the PRIMEs, it is hoped to adopt very similar techniques to support the GEC systems although this will be centred at Manchester. Long term, this will provide 2 support staff at Manchester and 1 or 2 at Rutherford. There seems little hope of reducing the number of support staff as the requests for help are non trivial and are not reducing in quantity. Help is often required for such tasks as requests to read magnetic tapes and other special facilities.
The User Services developed by ICF since its inception in 1975/76 may be classified under the following seven headings:
These are described in more detail below. The ICF also mounts and maintains applications software for a substantial section of the research community. It also provides training in its use and develops the software in response to user requirements. Further details of software available is given in section 6.
Potential users of the ICF are encouraged to have preliminary discussions with ICF management and technical staff before applying to use the facilities. As a first point of contact each installation has a manager who acts as the ICF management representative. Advice can be given on the local services available and on how to apply for them. Where a user requirement arises which cannot be satisfied from existing local resources, the user is passed to the ICF Management at Rutherford and Appleton Laboratories (RAL). Staff there can discuss the problems of providing the necessary services including those on Central Batch Computers, Microfilm Recorders, ICF terminal pool and the SRC network and communication facilities.
One of the major problems facing an applicant especially one with little or no experience of interactive computing, is the assessment of what is actually required to support a particular research project. This may include:
Once the potential user is aware of his requirements, he may make applications to use the facilities. There are two broad classes of application:
A SRC Approved (free access) B Use on repayment
Each is divided into smaller sub-classes:
In most cases the primary application form is the AL54 form which summarises all the required facilities and is accompanied by a technical summary of the computing aspects of the project.
In cases where approval is made through the normal research grant procedure the AL54 is appended to the RG2 or other paperwork and the whole is considered by the relevant subject committee together with comments from the ICF management. These relate to ability to provide the required services, including suitability of equipment, availability of computer resources, terminals, communications, ICF facilities and software.
When approval is obtained through any of the mechanisms mentioned above, an authorisation is set up detailing all of the equipment and facilities that will be provided for use by the project. Copies of this are sent to all relevant staff for action.
The authorisation which arises from approval to use the ICF is used by the various sections of the management at RAL and host sites to set up a new project. This consists of some or all of the following activities:
The basic resources of the computers are rationed per four week statistical period (SP). The ration is computed from the total number of Allocation Units and the length of the project. Installation managements have some freedom to vary this ration in response to individual and community requirements provided that the individual projects do not exceed their total allocation. To help spread the loading on each machine, a task carried out during evening, night, or weekend hours is only charged at one-tenth of its cost if run in peak computing time.
Regular accounts of each project's use are kept and are always available to assist supervisors to maintain overall control of their resources.
The principal machines are DEC, PRIME and GEC. Daily running is by SRC staff at RAL, or University staff for machines located in a university. All equipment is maintained under contract and regular machine management activities, such as disk dumping are carried out. The system software is always standard for a range of machines and is tested and distributed by RAL staff, usually via the network.
All machines operate on a 24-hour day, 7-day week basis, apart from scheduled stand-alone maintenance. Each has two or more demountable disk drives which offers the user population an average of 1-2 Mbytes of disk space per user. Each machine has slow peripherals including a Benson 1302 three-colour drum plotter and a 300 lpm lineprinter. Most machines have one or more local pools of terminals which are available to all users as well as links to more remote users. Most machines also provide dial-up modems to allow use outside working hours and access by remote users awaiting permanent connections.
The ICF runs a pool of some 650 interactive terminals and associated equipment. This equipment is used in two ways:
The distribution of terminals is illustrated in Figure 1.
Terminal equipment is maintained by a contractor with nationwide coverage except where the maintenance requires very specialised knowledge.
The ICF Management uses where possible Post Office leased data communication circuits and dial-up services.
The leased circuits are used for two purposes:
Dial-up services are used for low volume traffic where a leased line cannot be justified and high speed is not required or prior to installation of a leased line circuit. The availability of these services also allows users to access the facility out of normal working hours using portable terminals with acoustic couplers. Most sites have such a terminal which can be borrowed for short periods.
Funds are provided for the use of dial-up services, and for the installation of local communication links within a university campus, but where possible existing lines are used.
Each host computer site has some user support capability to deal with local user queries on a day-to-day basis. This is usually provided by the local manager although in the case of larger installations (Edinburgh, UMIST, RAL) separate staff have this responsibility.
All systems have an electronic mail facility used as the main method for notifying problems and also used to channel queries to central support staff at RAL.
Training courses for new users are run from time to time for particular machine ranges and documentation is provided to lead the novice user through the first stage of access to the facilities.
Each machine range has one or more user representative meeting(s) held to provide a channel of communication between the user population and the ICF Management. Such meetings are generally chaired by a member of the ICF Committee and thus provide user representation at the highest level.
This section illustrates the growth of the ICF user community over the past year. All accounting is done on the basis of individual user names and some accounts have many user names, so this has been a difficult exercise, and it is impossible to say exactly how many research grants involve use of the ICF and to what extent. The distributed nature of ICF means that accounting information is spread over many machines in different formats and, in the case of the upgraded sites, is difficult to obtain.
Figure 2 is a map showing the current state of the network. The various machines are listed at the foot of the figure.
Figure 3 indicates the growth in the number of registered users over the past year broken down by machine type. It also shows the number of active users in each statistical period. Comparing the two indicates that in an average period 58% of the registered users make use of the facility.
Finally, it breaks down the number of AUs delivered per period by machine type. It can be seen that over the past year the PRIMES have produced 8649 AUs, the GEC machines 6341 AUs, and the DEC 10s 17,138 AUs. In addition, other upgraded sites produced some 7960 AUs. The average number of AUs produced per period is 3075 and 63% of these have been taken up by SRC supported users.
Accounting figures for the upgraded MUM sites at Leeds, Oxford, Sheffield, Southampton and Swansea are less precise since nonstandard accounting software is used. (The data for the Nottingham and UEA sites are included in the PRIME figures since these sites run standard SRC software.) The available statistics indicate however that both the number of SRC users and the amount of time used by these users is steadily increasing.
In the past year we have dealt with about 320 research grant applications seeking ICF facilities. The number of applications made under the new Streamline procedure is also growing.
Applications software has been obtained from a variety of sources. For convenience, these have been divided into three main groups - software resulting from Special Interest Group activities; basic software which is in general not application specific; and other software which is available on only one machine range. Each contributor has been asked to describe any major impediments to current or future developments.
The work has been carried out by the University of Edinburgh and all programs run on the DEC-10 at ERCC. The major proposal was for the development of a new POP-2 system called WPOP. This has been under development for four years, and is still being improved. POP-2 is a highly interactive language system giving the user a total programming environment. This provides an editor, compiler, debug facilities, etc normally only available as modules. Comprehensive data structures and control facilities are supplied. There is also an associated source library of approximately 20 programs. (There are also two other POP-2 systems available but these are not maintained by the SIG.)
Three dialects of LISP, the list processing language system, are maintained; these are all incompatible but each system has its adherents. Stanford LISP 1.6 is the smallest and is used as a kernel for building larger systems. The CONNIVER system requires the MACLISP dialect. The Rutgers/UCI LISP is the cleanest and is recommended for general use.
Support for PROLOG, a language for programming in predicate calculus, will shortly be undertaken by the SIG. The LISP-based symbolic algebra system is also supported by SIGAI.
Effort used:
Problems encountered or envisaged:
Projects supported by the SIG are given below. Only a few are currently using WPOP. The old POP-2 system, PROLOG and LISP are also used by these groups.
User:
University/Institute | |
---|---|
3 projects | Edinburgh University |
2 projects | Sussex University |
1 project | Warwick University |
1 project | LSE |
2 projects | London University |
1 project | Exeter University |
1 project | Liverpool University |
The SIG has just reported to ICFC and the programme has not yet started. Major proposals approved are:
The work for (1) - (3) will be carried out at RAL under the programme of the Pre and Post Processor Section (see Section 6.1.7).
A programme of software development was proposed to the ICFC in June 1979 and accepted. The main areas of work were:
All the above were to be mounted on a range of machines within the ICF in order to provide access for the user community. The software provided so far consists of the UMIST and Cambridge control system design packages. The Cambridge package has been mounted on the DEC10 at UMIST and, of course, is available on the GEC4070 at Cambridge. The UMIST packages are only available on the DEC10 at UMIST.
Planned for the near future is the mounting of the ACSL simulation package on the PRIME machines at UMIST, Sussex and Warwick. User support will be provided by the SIG and by Warwick/Sussex staff.
Due to difficulties in finding staff with the right background of control engineering and software development experience, only one of the two programmer staff is currently in post. A second programmer has been recruited but his appointment has been delayed due to recent new Government rules regarding work permits.
Effort has been limited and progress slower than expected mainly to a shortage of available staff of sufficient calibre and background knowledge. We hope to rectify this shortly. We have also been slightly restricted due to problems of access to the DEC10 machine (remote line faults, machine non-availability) and late provision of terminal facilities. However, work on the subroutine library is now progressing well. Joint work with Warwick on a data analysis/identification package should also begin shortly. We hope that machine access will improve with the introduction of the UMIST PRIME 750 and improved communication and terminal facilities, which have been promised for this summer.
The major user groups are situated at UMIST, Cambridge, Sheffield, Bradford, Warwick, Sussex, Bangor, Bath, Cranfield, City and Kingston. A list of specific projects is not available.
The Electronic Circuit Design Steering Group was formed from the amalgamation of the Digital Circuit Design and Analogue Circuit Design SIGs - SIGCD and SIGDD. The SIGCD proposals were to obtain (or write) and support software for:
SIGDD requirements were for software for:
Both SIGs required support for the software so that users could be trained in their effective use.
The following programs have been mounted on SRC machines. They provide the best packages currently available to the DACD group, and they cover all of the SIGCD requirements except (4), and all but (2) and (4) of the SIGDD requirements:
Software | Description | Machine |
---|---|---|
ICAP | AC Analysis | PRIME |
ITAP | Transient Analysis | PRIME |
SP | Scattering Parameter Analysis | PRIME |
NAP | AC, DC, Transient Analysis | IBM, PRIME |
SPICE | AC, DC, Transient Analysis | IBM |
ASTAP | More flexible analysis program | IBM |
SCEPTRE | Considers effect of radiation | IBM |
CSMP | General simulation and analysis | IBM |
ANP | Frequency and time-domain analysis | IBM, Prime |
DDL | Digital Design Language | DEC 10 |
DIGSIM | Statistical Logic Simulator | DEC 10 |
ISPS | Simulator for ISP language | DEC 10 |
PCIRC | PCB Layout | PRIME |
HELP | Comprehensive information and help system | DEC 10 |
For SIGDD requirement (6), the GAELIC package for IC layout is also available on DEC10 and Prime computers. Support is provided by Technology Division at RAL.
Two courses on NAP have been held to train researchers from universities and industry on its use.
Effort used:
Software assessment / selection 2my Software development 2my Software maintenance 2my User support 1.5my
Problems encountered and envisaged:
University/Institute | Project |
---|---|
Bath | Nonlinear magneto-static problems |
Bristol | Load variation analysis on communications network |
RAL (SNS Project) | Transient and frequency analysis of large power supplies. |
Southampton | Circular transistor modelling. MOSFET multiplier configuration modelling. |
Birmingham | Various analogue electronic design problems |
London | Data flow architecture investigation |
The following programs have been made available:
Software | Description | Machine |
---|---|---|
GFUN | Integral operator 3D nonlinear magnetostatic package; some beam tracing possible. | IBM |
THESEUS | Interactive preprocessor for GFUN | Prime |
BIM2D | Boundary integral solver for linear Poisson equation in 2D. | Prime + AP120B |
PE2D | Differential operator solver for non-linear Poisson equation in 2D. Can be used for eddy current calculations and allows some beam tracing. | Prime |
BIM3D | 3D version of BIM2D | IBM |
TOSCA | Differential operator solver for nonlinear Poisson equation in 3D. | IBM |
PE2D is being developed to include generalised units; infinite elements; fitting and editing of potentials for smoother fields; anisotropy; and hysteresis. In addition, an advanced post-processor is being developed under a contract with Infolytica Ltd. The TOSCA and BIM3D programs are also being interfaced to THESEUS.
Effort expended:
Package support 4.5my Secretarial work (meetings, courses) 1.5my
University/Institute | Project |
---|---|
Cardiff | Transformer limitation |
Bath | High homogeneity study Homopolar motor |
Aberdeen | Electrostatics (hazards) |
Southampton | Electrostatics (hazards) |
Leeds | Stepping motor |
Southampton | Magnetic levitation for wind tunnel |
Bangor | Magnetic levitation |
Imperial College | Machine problems |
Southampton | Machine slot study |
Surrey | Particle deflection magnet |
Bristol | Cathodic protection problem |
Culham Lab/RAL | In-house projects |
The Finite Element Steering Group was formed following the joint proposal by the two Special Interest Groups to ICFC in May 1978. FESG's major objectives have been to support two FE packages, following an evaluation exercise; to develop an FE Library for the large community of research workers who wish to develop software; and to provide interactive pre- and post-analysis tools to help overcome the problems of data generation and validation, and output analysis.
For those researchers requiring robust, well-developed software the ASAS and BERSAFE programs have been obtained for use on the IBM batch computers. ASAS is a large, well-documented system with a good user interface. It provides linear static and modal analysis capabilities. The BERSAFE suite of programs can only handle medium-size problems, but has a wider analysis range, including plasticity, creep, fracture mechanics and transient heat transfer.
The Finite Element Library is not a package: it is aimed at those users who wish to develop new analysis techniques. It is a two-level library, Level 0 being a set of subroutines to provide most of the basic operations needed in constructing an FE program. Level 1 consists of a set of programs showing how the Level 0 library can be used to solve a wide range of problems in both structural analysis and fluid mechanics. The Library is available on the GEC, PRIME and IBM machines. Three courses have been held to train researchers in its use.
The FEMGEN interactive graphics mesh generator has been obtained on Prime and GEC, interfaced to the available FE software. This program is well-liked by the users and provides an excellent tool for the pre-analysis phase.
Two general FE User Meetings have been held, and also the first in a series of one-day specialist meetings called FE Colloquia (on Fluid Mechanics). The user population comes from a wide background, since FE is now being recognised as a general technique for solving partial differential equations. The structural analysis area is still the best-developed however.
Effort expended:
User support 2 my FE package evaluation 1 my FE package maintenance 1.25 my FE library development 2 my Pre- and post-processors 1.25 my
Major limitation:
Lack of post-processing software. This requires effort which currently is not available.
University/Institute | Project |
---|---|
City | Instrument transducer modelling. |
Kingston Poly | Fracture mechanics |
Open | Plastic zones at end of notch |
Birmingham | Dynamic and acoustic behaviour of industrial machines. |
Heriot-Watt | Forced vibration of electric motor casing. |
Bath | Electromechanical stressing in electric motors |
Swansea | Infinite elements |
Robert Gordon's | Drilling fluid flow |
UMIST | Water movement in porous masonry |
Newcastle Poly | Structure of thin films |
Liverpool | Tidal flow in estuaries |
Surrey | Analysis of trombone vibrations |
Glasgow | Ultimate load design methods for plates and walls. |
Preston Polytechnic | Lubrication of bearings |
Bath | Diesel engine design |
TRRL | Tunnels and underground pipes |
The aim of this group is to obtain for ICF users general-purpose interactive pre-processors (eg, geometric modellers, mesh generators) and software modules for use in constructing interactive post-processors (eg, database management, high-level graphics, command decoding). The group has been in existence for less than a year. Software available shortly will be DBOS (a library of Fortran-callable subroutines to perform data management for applications software) and SIMPLEPLOT (a library of Fortran routines for producing good quality annotated graphs quickly and easily). Some of the proposals of the CAAD SIG will be implemented by this group. (See Section 6.1.2)
Effort expended:
Software planning, design and development of man-machine interface software 0.8my Acquisition and installation of software 0.3my Participation in geometric modelling description standards 0.2my
This section describes software which is not specific to anyone applications area, but is of wide interest. Each of the major ICF machines (DEC, GEC, PRIME) are supplied with the NAG Library as a standard numerical software library. The other major software groups are Database Management/Information Retrieval systems, Graphics software, and Statistical packages, which are described in the following sub-sections.
Three major items of software are described here - STATUS, PRIME DBMS, and MIDAS. All of these are available only on the PRIME 750 at RAL; however it should be noted that the DBOS system is being provided on a GEC machine under the Pre- and Post-processors programme.
STATUS is a full text information retrieval system with efficient online retrieval facilities. Half of the funds for its purchase were provided by ICF. There are a number of users, mostly at RAL.
PRIME DBMS is an implementation of the CODASYL Data Base Task Group 1971 report, with a few extensions. It provides generalised data management capabilities for creating and maintaining structured databases. Both hierarchical and network type data structures are possible. A Fortran callable interface is provided.
MIDAS is a data management system which can be used to set up and maintain large indexed sequential files. Up to 20 indexes are allowed for a MIDAS file and keys can be of various types. MIDAS is accessible via Fortran, and is suitable for applications where the quantity of data is large but the inherent structure is not complex enough to warrant use of a DBMS.
ICFC's recommendations at its November 1976 meeting were that:
FINGS was implemented as the minimal graphics package, with device drivers for Tektronix 4010 and 4014, .and the FR80. A driver for the Benson 1302 plotter has been added more recently.
GINO-F and GINOGRAF were purchased and implemented on the PRIME 400 (only the PRIME 300 version was available). New device drivers, most written internally, for Tektronix 4010, 4014; Imlac PDS4, Dynagraphic 3205; Sigma GOC T5600, T5670 (monochrome), 5664 (colour); Benson 1302 plotter; FR80; and Hewlett-Packard 2648A are now available. A Hewlett-Packard 7221B desk-top plotter device driver is under development. Little work was done in mounting GPGS-F as it soon became clear that the manpower available would only allow support for one major graphics package.
Effort expended:
Software development 4.5 my Software maintenance 1.0 my Monitoring of new graphics hardware 0.75 my User support 1.25 my
A major problem (which was in fact foreseen), was the inability of GINO-F to provide support for the capabilities of more modern displays and, in particular, raster displays. Efforts have been made to alter GINO-F to provide limited support. However, at some stage it will be necessary to change to a more modern system such as the proposed GKS.
The Rosenbrock report identified interactive statistical software as a requirement, but no formal commitment in this area has been made by ICFC. However, in support of some local RAL requirements, some software has been made available. The most widely used statistical package, SPSS, has been mounted in both the standard batch and more limited interactive versions on the PRIME. This has been used in work for the National Committee of Computer Networks and a survey for the Instrumentation Committee. SPSS has also been used to monitor use and response characteristics of the PRIME computers. SCSS, the fully interactive version of SPSS, is available on the ERCC DEC10 on an own risk basis. Both GLIM and GENSTAT have been mounted on the RAL PRIME, mainly due to interest by a group at Leicester Polytechnic.
Signal processing/time series analysis software is, of course, available on the upgraded University MUM at the Institute of Sound and Vibration Research at Southampton University. Cranfield's SPAG system is now available on the ICF GEC MUMs.
(All ICF machines by default have the NAG, GINO-F and GINOGRAF libraries.)
Both UMIST and ERCC have the following languages available:
AID ALGOL BASIC BCPL BLISS-10 FORTRAN MACY-11 PAL-10 SAIL SIMULA
The RUNOFF text layout package is also mounted on both machines.
Further software items available at ERCC only are:
APL BLISS-11 FASBOL IMP REDUCE-2 LAYOUT text layout system CALCOMP graphics library SMOG graphics library CSSR crystal structure information retrieval system
At UMIST additional items are:
ALGOL COBOL68 FAIL PALX11 PASCAL EISPACK Argonne eigenvalue/eigenvector extraction routines POWDER Refinement of crystal structure using profile analysis
The following languages are available:
APL BABBAGE BASIC BCPL FORTRAN
The GEROFF text layout packages is also mounted. A wire wrapping design program, WIREWRAP, is mounted on the RAL machine.
Languages available are:
ALGOL BASIC BCPL FORTRAN MICROSIM PASCAL PMA
The DUCT geometric modeller has been obtained for the RAL PRIME 750. WIREWRAP is also available on the Chilton machine. The PRIME RUNOFF system is widely used for documentation.
The Nottingham PRIME also offers SAMMIE, a 3D modelling, ergonomic and robot workplace evaluation system and VIEW, a generation and selective viewing program based on GINO-F pseudo-code pictures.
At the University of East Anglia, the following software is additionally available:
BUILD geometric modeller ALTRAN symbolic manipulation language RATFOR structured FORTRAN pre-processor SPICE electronic circuit emulator AESOP database system
This machine was bought for evaluation as a cheap (approx £50k) high speed attached processor for the PRIME A machine at RAL. The aim was to give more cpu power on an interactive timescale. Initially the machine was difficult to use, the only modes being the Maths library or a difficult assembler. In October 1979 a compiler (APFORTRAN) and link-editor (APLOAD) were installed, and usage has steadily increased since then. This software, which runs on the PRIME, is slow and cumbersome; compilation is normally done in batch mode. A significant amount of work has been done on the software at RAL , particularly by modifying the input/output and library searching. A speed gain of a factor of 10 has been achieved over the original software in link-editing.
Using APFORTRAN a speed gain on execution on the AP120B over a PRIME 400 of between 4 and 9 times is achieved. If the assembler is used, gains between 10 and 30 times a PRIME 400 are normally achieved. The drawbacks to current use are the 38-bit word length (inadequate for some applications areas); lack of COMPLEX and DOUBLE PRECISION; and the small program memory which allows a maximum of between 400 and 600 lines of FORTRAN.
Effort expended:
User support 1 my APFORTRAN, APLOAD modifications 1.25 my Maintenance 0.25 my
User groups are:
University/Institute | Project |
---|---|
RAL (SIGEM) | Two production codes running |
Reading | Semi-conductor simulation |
Bristol | Algorithm development in fluid dynamics |
York | Crystallography |
Leeds | Chemical engineering |
Oxford | Image analysis of electron microscope data |
Salford | Fluid dynamics |
Sheffield | Control Engineering |
The Sheffield Interdata 8/32 offers the FORTRAN, BASIC, RT1/2 and CORAL66 languages. Applications software available includes
SPAID | general purpose package for analysis and identification of data. |
SCADS | control system analysis and design suite. |
ICPDS | interactive continuous process dynamic simulation program |
PSI | interactive block-structured simulation program |
Relay package | subroutines to obtain solutions for multi-phase oscillations in feedback systems |
Microprocessor simulators and cross-assemblers for Motorola M6800 and Texas 990/4 | |
Packages for various numerical analysis areas and graphics. |
The Southampton (ISVR) PDPII/45 offers a wide range of software in four main groups - analysis and processing modules; graphics; file arithmetic and manipulation; and data acquisition and input/output.
The UC Swansea PDP 11/45 has the Grapho graphics system, and the RUNOFF text processing package. Extensive finite element research work is done via this machine.
At Leeds University, the VAX 11/780 offers software in the following areas:
The following table gives a breakdown of staff employed at RL and at remote sites. The figures are targets which it was hoped to achieve in the current FY. It was assumed that we would succeed in using the actual manpower of 46 allocated within RL, but an SRC-wide interruption of recruitment (see 7.2.1 below) will now make this impossible. The Operations and Systems support staff for PRIME and GEC systems is, in general, made up of 0.5 my for operations and 0.5 my for systems support per remote site.
Remote Support | ||||
---|---|---|---|---|
Function | RL | ERCC | UMIST | Other |
DEC10 Systems/User Support | 6.0 | |||
PRIME + AP120B Systems | 2.5 | 2.0 | ||
PRIME User Support | 2.0 | 5.0 | 4.0 | |
GEC Systems | 2.0 | |||
GEC User Support | 2.0 | 5.0 | ||
PRIME/GEC Technical Management | 1.0 | |||
Upgraded MUM Systems/User Support | 8.0 | |||
Comms Software (general) | 1.0 | |||
Real-time connections/SUM | 1.0 | |||
Operations | 2.0 | 2.0 | 9.0 | |
Management/Committee Support | ||||
General management/Committee support | 2.0 | |||
ICF Sites Management/Resource Allocation | 2.0 | |||
Terminal Pool Management | 2.0 | |||
Admin and Secretarial | 5.0 | 1.0 | 0.5 | |
Grant Assessment | 1.0 | |||
Databases | 1.0 | |||
Application Software Development/Support | ||||
Management | 1.0 | |||
AI | 0.5 | 2.0 | ||
CAAD | 0.5 | 1.0 | ||
Control | 2.0 | |||
Finite Element (library and packages) | 4.5 | |||
Pre/Post Processors | 1.5 | |||
Circuit Design | 5.5 | |||
Electromagnetics | 3.0 | |||
Graphics | 2.0 | |||
AP120B | 1.0 | |||
TOTAL | 46.0 | 9.0 | 7.5 | 31.0 |
Loss of highly-qualified staff, generally to industry, is a serious problem: recent losses of systems programmers, user support staff and a mechanical engineer in FE applications will be particularly hard to replace. As a result there will be a tendency not to use the full allocation of my in a given year. Permanent staff may only be recruited into SRC at HSO level or below without special cases being made. Consequently, it is very difficult to recruit engineers with any degree of expertise in a particular discipline.
Advertisements to replace lost staff recently had to be withdrawn in an attempt to reduce expenditure in FY 79/80 below levels precisely allocated by the Engineering Board. This drastic step will have particularly serious effects in production of SIG software for Circuit Design and Finite Element applications.
There will be a total of 19 GEC machines (Table 1) supported by ICF staff by early 1981, of which only 12 were originally purchased by ICF. It has been our policy to offer support of ICF systems software on other SRC and NERC machines. NERC pay for the work while other SRC projects currently provide neither funds nor manpower. However it is now a full time task for one man merely to maintain the operating systems at the sites and we may have to review the situation'. It will not be possible to reduce the number of staff working on these machines below two. There is also a need for a GEC 4000 system devoted purely to system development and generation.
Development of an archival system is now urgently needed as the discs are becoming full on several sites.
The major piece of work still to be done is to complete the networking software in order to:
Security is a relatively weak feature of the PRIME operating system and this may require some improvements later.
The movement of systems support and user support tasks to UMIST has not yet provided benefits but is already absorbing effort in liaison and management: this must be continued as rapidly as possible until at least some reduction in total RAL effort is achieved. If there is a continuing need for system development, thought will need to be given to the provision of a PRIME system purely for system development.
There are difficulties at the moment in handling
All areas involve the need for considerable technical knowledge as well as administrative skills: the problem appears to lie more in finding and keeping staff with appropriate qualities, rather than in the numbers required. The difficulties here may be an indication of the considerable management problems which more widespread SRC/NERC/CB networks may bring in the future.
It is particularly difficult to find and keep engineers with programming skills in the digital circuit design area, and only a little less so in stress analysis.
The end of Phase 2 represents a distinct watershed in the development of ICF. The programme for installing Multi-User Minicomputers handling 6-8 concurrent users will end at latest in 1981 - indeed cuts in Engineering Board budgets already envisaged for that year may even have the consequence that new Multi User Minis cannot be provided even at sites whose cases have already been accepted. We see two main needs for provision of hardware beyond 1981:
As anticipated in the Rosenbrock Report powerful single user machines are now beginning to appear. Both the Three Rivers PERQ machine and the recently announced Apollo machine, now offer to a single user similar power, main memory and filespace as a complete MUM. A high resolution A4 display is also included, together with various interfacing options within a total cost of about £12k. It is fervently to be hoped that one of these machines or something closely resembling them, will soon become available from a UK manufacturer.
Nevertheless, the dangers in SRC making SUMs their principal provision for interactive users, look very similar to the dangers which Rosenbrock was aiming to circumvent when the ICF was first proposed. These are, that individual users will revert to doing all their own development work on their own machines, generating software for their own sole use and collecting and keeping their own data sets, all of this being grossly prodigal of research workers time. There is also the essential requirement for SUM users to be able to reach major batch facilities.
For these reasons ICFC is recommending that no widespread purchase of SUMs should begin until hardware and software has been produced to permit networking that is fast, cheap and completely proven. At that stage SUMs can be linked to one another and to fileservers - possibly existing MUMs - via local area networks, and can use long range networking to exchange programs and data with other sites. We see the work of the Joint Network Team, and especially the introduction of campus networks as a crucial pre-condition to widespread installation of SUMs.
The present ICF network comprising MUMs linked to central SRC batch facilities has satisfied two classes of user:
First, the engineer needing good local interactive graphics for design of systems for which the analytic calculation is relatively simple eg, the design of multi-variable control systems.
Second, the engineer who wishes to model interactively the boundary conditions of very complex field problems, eg, in electromagnetics or stress analysis. An interactive terminal is used to generate a satisfactory model and mesh which is then passed over the network for the main calculation to be performed by a central number cruncher in batch mode.
In the second case, however, the result of the job will not be returned to the interactive machine for post processing until many hours have elapsed. Thus the procedure is suitable for analysis of a single model, but not for true interactive design leading to the synthesis of a new model.
Neither the existing MUM systems nor a future network of SUMs can provide for the user who requires the full power of a massive number crunching machine, for example, a Finite Element problem requiring 20,000 nodes, if all this power is needed within an interactive time frame. Yet this sort of provision is absolutely necessary if the power of computers is to assist the design, not just the analysis, of complicated parts and systems involving solution of field problems.
The ICF Committee believes that there is a limited subset of SRC grant holders who could make the special case to be provided with such an expensive facility and that they should be encouraged to do so. Existing central machines such as the IBM or the Cray cannot provide this, since even if spare capacity were available, their architecture is not such as to allow them to be used efficiently in providing an interactive service at the same time as fulfilling their main batch function.
For these reasons it is believed that ICF should provide a single interactive high power processor with a performance of about five 360/195s, its use being carefully restricted to a small number of grant-supported users who could make the case outlined above to their Subject Committees for specially heavy interactive computing at a high notional cost. Provision of £2.5M has been made in the current 5-year Forward Look for this purpose, although it is too early to attempt to identify a suitable machine.
It is believed that the best strategy for the next 4-5 years will be simply to continue vigorously the present programme whose results so far are outlined in Section 6, above. It is expected that more Special Interest Groups will be formed as a result of pressure from individual Subject Committees and, although it is too early to say how many, the Forward Look has assumed that a further two SIGs will be set up within the next two years. Programming effort in support of these has been included.
Further applications software work should be aimed not just toward MUMs, but also toward Single User Minis and the Interactive High Powered Processor.
The existing ICF network of MUMs, together with the machinery for managing it and allocating resources on it, will be handled by six functional groups now established within the Computing Division, RAL. All the existing workstations to central batch facilities, including enhanced workstations, will be managed in the same way and funding for them and for the ICF MUMs should be from a central SRC Computing Committee.
The feasibility studies and the initial installations for Single User Minis including the provision of networking for them (see 8.1 above) should be undertaken by a single project team in Computing Division, tightly related to the group continuing the future Applications Software work.
This team should be funded and monitored by the same Engineering Board Committee for Computing that is charged with development of the Applications Software. The detailed arrangements for integrating the existing ICF network with the remainder of SRCs computing activity must depend on recommendations to be made in the final report of the Computer Review Working Party and it is recommended that firmer plans for implementing Phase 3 of ICF should be drawn up after that report is available.