Opinions on standards never go away; the article on SGML is of special interest and the offer of a tutorial on this subject is free to readers. Other training opportunities abound this month. Last month's newsletter included the name of your EASE Contact, who will help with advice about the EASE programme. Articles by you, our readers, are important, so if you have anything of interest please write about it to share with us all. If you are reading this newsletter from a notice board or from a circulation system but would like to receive your own copy, I would be pleased to add your name to the mailing list - just let me know.
At around 3 pm on Friday 30 March 1990 Professor Alistair MacFarlane, Principal of Heriot-Watt University, pressed the return key (see photograph). This switched off the last Interactive Computing Facility (ICF) Multi-User Mini (MUM) - the Prime 9955, better known as RLPA, at the Rutherford Appleton Laboratory.
Around fifty invited guests attended the ceremony in the Atlas Centre to mark the end of what has been a major success story for the Engineering Board. Key people involved in the ICF throughout its fifteen year life came together to discuss the past and renew old acquaintances. Brief presentations reflecting back on the ICF were given by Dr Gooff Manning (Chairman of AMT), Professor Howard Rosenbrock (recently retired Head of the Control Systems Centre at UMIST), a video recorded message from Mr Peter Davey (Oxford Intelligent Machines Ltd) and Professor MacFarlane, before the last MUM was officially switched off.
In a series of articles over the next few issues of this Newsletter I will be summarising, with the help of a few photographs, the fifteen or so years of the ICF. If any readers wish to provide me with input for these articles - any interesting photographs would be welcome - or perhaps an article reflecting their own experiences with the ICF, please contact me at your earliest convenience.
This article aims to inform the engineering community of the current status of Advanced Research Computing at the SERC Daresbury Laboratory, and invites the community to use the facilities described herein for evaluation and development purposes. This invitation is offered in view of the Laboratory's role as an Associated Support Centre for the SERC/DTI Transputer Initiative.
Before considering the parallel processing hardware, we outline briefly the general computing and support structure at the Laboratory. Daresbury now provides a UNIX-based distributed computing environment, comprising a range of mini-supercomputers, super-workstations and, the subject of the present article, parallel processors. All these facilities are available to the academic community through JANET. The main work-horse is a dual processor Convex-C-220, with additional computational power provided in-house by an FPS M64/60 attached processor. Superworkstations used in areas such as molecular modelling include the dual-processor Stardent Titan2, Apollo DN10020 and Silicon Graphics 4D/220GTX.
The Advanced Research Computing Group (ARC G) at Daresbury is involved in both parallel algorithm design and the development of parallel processing tools, such as the FORTNET harness. Its major task, however, is to ensure the successful migration of application software from the more traditional environments associated with the National Centres at RAL, ULCC and MCC, to the novel architectures described below, and to assist the community in the cost-effective scientific and engineering exploitation of this hardware. The ARCG is very much project oriented, with a strong overlap between its work and that of the CCPs (Collaborative Computational Projects). The ARCG also hosts workshops and symposia organised under the auspices of the Computational Science Initiative's (CSI) Parallel Computer User Group, the most recent of which, held on March 6 at Daresbury, presented details and results of work around the Intel iPSC/2.
Parallel processing activities at the Laboratory are thus focused on the IPSC/2 hypercube from Intel Scientific Computers, and a rather modest Meiko Computing Surface. The second generation Intel concurrent processor comprises 64 multiple instruction multiple data (MIMD) nodes placed on a hypercube with direct-connect channels. The hypercube was installed in October 1988 as a 32-node machine, and was upgraded in June 1989 through the provision of an extra 32 nodes, and the introduction of the Concurrent I/O system. The former upgrade was funded through the collaboration between Daresbury and ICI in Computational Science.
Each node of the iPSC/2 consists of an Intel80386 processor with 4 Mbyte of memory. Enhanced floating point performance is achieved through a Weitek 1167 SX scalar accelerator (0.6 Mflop 64-bit or 1 Mflop in 32-bit arithmetic). Vector processing on 32 of the nodes is available through a VX-vector accelerator (6.6 Mflop in 64bit or 20 Mflop 32-bit). The vector board has 1-Mbyte associated memory which can be addressed, making a total of 5-Mbyte node memory.
Inter-node communications are handled through the DMA (Direct Memory Access) node-controller, which provides a maximum 2.8 Mbyte/sec asynchronous communications network on each of eight bidirectional channels, with seven node-channels available for the network and one for I/O. In contrast to the first-generation iPSC/1 the cube may, from an application viewpoint, be considered as an ensemble of fully connected nodes, since there is now little more delay in long-distance messages than in neighbour-neighbour transfers.
The cube itself is controlled by a front-end processor, the system resource manager (SRM), which is an Intel System 310 microcomputer with 8 Mbyte memory. The SRM is linked directly to the cube though direct connect hardware, with access to other UNIX workstations provided over Ethernet and TCP/IP networks. With a UNIX System V operating system, the SRM compiles, links, debugs, allocates and communicates with the hypercube. In order to reduce loading on the SRM, Intel provide 'remote hosting' software which enables users to run programs from other UNIX machines. Remote hosting is at present possible at Daresbury from either Sun workstations or the Convex-C-220. This capability of dividing the workload between a UNIX engine and the multicomputer provides a valuable resource given hosting by a powerful UNIX processor such as the Convex-C-220. It enables a balanced solution whereby the well defined highly parallelised compute-intensive parts of the application are allocated to the cube, while any significant serial component, together with pre- and post-analysis associated with the application are performed on the host. This dual approach has been used to advantage in various application areas.
Each iPSC/2 node carries its own multi-tasking operating system, called NX/2 (Node eXecutive/2). An effective subset of UNIX, NX(2 provides multiple tasks with private address space, dynamic memory allocation, UNIX I/O, round robin scheduling of processes, and supports the DECON parallel debugger.
The UNIX common object file format is supported and routines may be called from Fortran. NX/2 provides the program's interface to the Direct-Connect communication network, with support provided for the sending and receiving of both synchronous and asynchronous messages.
In addition to the hypercube-connected set of compute nodes, the iPSC/2 addresses the requirements of high capacity, fast access mass storage through the Concurrent File System (CFS), an I/O subsystem that features an attached set of I/O nodes. Each I/O node has full access to the hypercube interconnect of the computational nodes, has 4 MByte of memory and a SCSI bus which connects to one or more high capacity drives. The iPSC/2 I/O subsystem at Daresbury comprises two I/O nodes at present, each with two 574 MByte disks, permitting fast parallel random access to 2-GByte of data from any of the hypercube nodes.
The software environment surrounding the hypercube is hosted by the SRM and may be remotely accessed from any UNIX workstation. This software comprises a range of facilities. Compilers for Fortran-77, C, and concurrent Lisp are available, with VMS and Berkeley UNIX 4.2 Fortran extensions. Vector processing tools include the V AST2 vectorising pre-processor. V AST2 accepts Fortran-77 and -8X constructs and generates VecLib calls. It vectorises DO and IF loops, provides help and warnings, optimises performance, and provides a high-level interface to VecLib, a library of micro-coded vector routines, comprising the majority of the level-l BLAS. The CFS software consists of specialised processes which run on the I/O nodes, and file management code which runs in conjunction with the application on compute nodes. The UNIX programming interface provides a single file system view, beneath which CFS manages parallelism and concurrent access to files. Support is provided for both synchronous and asynchronous Fortran I/O. Finally, a parallel symbolic debugger, DECON, designed for message-passing analysis and repair is provided for code development on the hypercube.
The Daresbury installation will be upgraded still further in April 1990 with the procurement of a 32-node iPSC/860 multicomputer. Each processor node contains the new Intel i860 microprocessor chip which, with 8 MB ytes of local memory, is clocked at 40 MHz and has an effective peak rating of 40 Mflop for the 64-bit arithmetic characteristic of scientific and engineering computation. In addition to the iPSC/860 from Intel Scientific Computers, the i860 forms the basis of range of new machines, including those from Alliant (FX2800), Stardent (500) and Meiko (MK086). Developments of the i860 multicomputer are at the heart of the DARPA-funded Touchstone Project, which includes within its set of milestones the development of a prototype 2000-node machine and 100 MBytes/sec communications capabilities.
It is perhaps also useful at this point to assess the performance likely to be achieved on the RISC microprocessor, as distinct from the peak figure quoted above. While crucially dependent on the quality of the Fortran compiler, our own initial assessment would suggest a ceiling of 4-6 Mflop in scalar code and perhaps 10-20 Mflop in vector code characteristic of the level-3 BLAS. What has to be of benefit to the end user, however, is the substantial effort being put into compiler development for the i860. While the only serious compiler for the 386based iPSCI2 was that due to Green Hills, Fortran developments around the i860 include those due to Alliant (through the Intel PAX agreement), Compass Inc, Green Hills (as running on the current iPSC/860 systems), the Portland Group, Multiflow and others.
Our own benchmarking of the machine suggests that, with the availability of an optimised Fortran compiler, the i860-based multicomputer will prove at least an order of magnitude more cost-effective in computational science than the recognised mini-supercomputer leader, the Convex C-2.
We conclude by reminding the engineering community of the availability of the concurrent hardware under discussion, and invite those interested to contact either Robert Allan or Martyn Guest at Daresbury to acquire a user ID and relevant documentation. Copies of the proceedings of the latest workshop, together with a directory of ARCG activities are available on request.
Users who would like to submit jobs to the Atlas Cray computer from their Sun workstations (if they have Blue Book FfP) may be interested in a short book of examples of running programs on the Cray batch system via the IBM Jobmill.
The following topics will be covered: compiling a Fortran program with and without a dataset; returning data output via FfP; returning graphics metafile output via FTP; using library programs; saving compiled binary programs on the Cray; accessing and running compiled binary programs on the Cray; accessing compiled subroutines from a program; storing and accessing programs and data on the IBM front end; file management on the Cray and the IBM.
The Cray operating system COS is shortly giving way to a variety of UNIX called UNICOS, and the book, Examples of Submitting Jobs to the Atlas Cray from a Sun Workstation, should be in circulation soon after this event.
What is SGML, or Standard Generalized Markup Language, to give its full title, and what relevance does it have to the academic community in the UK?
Well, firstly SGML is a Standard. It is an international standard developed over many years within ISO and finally published in 1986 [SGML]. The standard is now in use throughout the world. SGML is also Generalized. It has not been developed to satisfy anyone particular market place. This has its advantages and disadvantages, but the primary advantage is that it does not just apply to your particular area, or more importantly, just to someone else's particular area with whom you wish to communicate. SGML is Markup. It is additional clear-text characters added to a data stream which provide additional information about that data stream. Finally, SGML is a Language. More correctly, it is a Metalanguage - a language for writing specific markup languages in.
However, the most important aspect about SGML is not held in its title at all - SGML provides for a rigorous model to be defined for the structure of the data-stream to be processed. That model definition or Document Type Definition (DTD) allows users to define precisely the model they wish to use for every document. For example, if the first few characters in a data stream were <article> one may expect that what follows abides by the DID for an article. Because of that rigorousness, it is possible (and practical) to produce a wide variety of software which could process such documents or data streams. The obvious item of software would be a formatter, but also database and information retrieval software (Find all articles with author = P A Ellison and subject = SGML?) and more esoteric software like algebraic analysers which would extract maths from the marked-up document and then process it, or even include it into computational programs.
What relevance does SGML have to the UK Academic Community? I believe it has a lot of relevance. In a short article like this one, I can best describe that relevance by specific examples (and at the same time give an insight into the range of disciplines to which SGML can be of value).
First, the Advisory Group on Computer Graphics has adopted SGML for the documents it wishes to distribute to universities and research councils. These documents will consist of user guides, examples, manuals, etc with the viewgraphs and other graphical information encoded in Computer Graphics Metafile (CGM) within the SGML documents. They adopted SGML as their base standard because user sites had demanded information in a variety of forms - in TeX or troff, already printed, in an editable form, or even suitable for A5, suitable for A4, single column, double column, etc. So, by adopting SGML (and, initially, converting to TeX and to troff forms in- house) they have a single format to which all their disparate authors may write. Eventually documents would be distributed in SGML form if and when sites obtain their own SGML-based formatting systems.
Secondly, the Euromath Project is a project funded by the Commission of the European Community to provide value-added network facilities for mathematicians in universities throughout Europe. The intention is to provide a user interface which will allow the mathematicians to create their documents which include maths on screen, to email them, to merge them, to perform algebraic analyse, to compute them and ultimately to publish them without the multiplicity of re-keying that goes on at the moment. SGML will be the basis for the document and mathematical expression definitions even if the mathematicians should! will never see that - they will use a WIMP interface.
As a final example, the Text Encoding Initiative with its UK arm at the University of Oxford needs a means whereby it can convert texts (some extremely old) into digital form yet retain all the information concerning those texts, including textual structures and layout structures if necessary. They are adopting SGML as it provides for concurrent structures to apply to the same data stream.
These three very different projects all use SGML. Because SGML documents can carry their definition with them, sites wanting to process documents from these three projects and from the many others that are starting, will only need one set of software. To investigate the current batch of commercial software and to make recommendations to CHEST concerning purchase, as well as to provide a focus for SGML activity nationally, a small project is being proposed to be sited at the University of Exeter. This will be reported on further as the project develops.
A joint EASE/Computer Board-sponsored Awareness seminar on SGML will be held on Friday 15 June 1990 at the Manchester Computer Centre, University of Manchester, commencing at 10.00am.
The presenter will be Paul Ellison, University of Exeter, and attendance is free. Further details, registration forms, etc may be obtained from: Mrs Mary McDerby Computer Graphics Unit University of Manchester
The Computing Facilities Committee (CFC) has set up a Club and is inviting participation at an open meeting to be held at UMIST on Monday 4 June. The aim of the Club is to support Data Modelling and Data Management in the academic engineering community. The scope of this Club will include that of the former Community Club on Data Exchange.
The following is a summary of the proposal for a work programme for the Club, to be discussed in more detail at the open meeting. This should be seen as a suggestion only as participants at the open meeting (see below) will be encouraged to help define the work programme.
The Community Club will be formally established at an open meeting which anyone with an interest in engineering data modelling or management is encouraged to attend. The meeting consists of two parts, the fIrst one is devoted to presentations on subjects which are relevant to the subject of the club, the second part will be a discussion on the work programme of the Club. At this meeting the views of the community will be sought on the program described above, and also additional proposals for the work programme can be put forward. An active involvement of the community is seen as essential to the success of the Club. The open meeting will take place in Manchester (UMIST) on Monday 4 June.
Below is the preliminary programme of this meeting
If you have an interest in the work of the Club but you are not able to attend the open meeting, then please send us your name and address and we will keep you informed of developments.
Yugraph 90 is a computer graphics conference with a strong engineering flavour. About 50 papers will be presented with most of the authors coming from Eastern Europe and Italy. Most of the CAD/CAM areas are represented but there is also a strong emphasis on the theory of solid modelling, curves, basic algorithms and user interface techniques. Copies of the programme are available from me.
An EEC-sponsored residential course on Computer Algebra for Engineers and Scientists will be held from 17-21 September 1990 at the University of Bath. The course will be led by J H Davenport, author if the text book Computer Algebra and Professor of Information Technology at the University of Bath, home of Britain's largest computer algebra research group. The lecturing and demonstrating staff are drawn from the University of Bath, Queen Mary College (London) and the University of Nice. The five-day course will comprise lectures, demonstrations of software and hands-on experience of these systems on the University's network of workstations.
When you are looking into new ways of tackling problems, it is inevitable that new tools and new types of tool will crop up. Sometimes it will be obvious which of these are useful and, if you are already familiar with them, which may be appropriate for you. More often, however, you do not know what is available. Then you could either spend valuable time re-inventing tools and facilities which already exist and which you could have used immediately, or you know about them but do not know whether they are really what you need. Working out whether a tool is useful, let alone best, for your task can be a significant piece of work in its own right.
We can help you with this problem in several ways. First, we have used many of the most commonly available tools and know about lots of others, so we can give you advice. Secondly, we will be giving a short course on this very topic at the end of May (details below). Finally, we can give you access to particular tools and help you try them out on your problems.
The last of these means that you can conduct feasibility studies before submitting a research proposal so as to make it stronger and more focused, or as the first step of an accepted grant before committing to the purchase of a tool when you do not know if it is altogether suitable for your particular purpose. In order to do this, you need to get your hand on the tool, you need someone to guide you in its use, and we can help you with both of these. Here, at AlAI in Edinburgh, we run the Knowledge Representation Systems Trials Laboratory. Visitors to KRSTL can spend time using any of a wide range of specialised tools with help from people who have used the tools on real problems. A visit can last any length of time from one to ten weeks.
The EASE Programme supports this facility which is free to approved visitors. Sometimes they are also able to provide travel and subsistence expenses, though people are always encouraged to build funding for this purpose into their grant proposals.
Please contact us if you would like more details of this Visitor Scheme.
This one-day course is intended to provide you with an awareness of what it takes to select an appropriate development tool for a knowledge based system (KBS) application. We will be investigating ways of classifying tools and tasks. We hope to make you aware of the issues involved in mapping the features of a particular application to the facilities that are provided by KBS tools. Hardware, software and non-technical factors that influence the choice of tools will also be considered.
A common thread running through the course is the acknowledgement that the selection of KBS development tools, in keeping with the nature of artificial intelligence programming, has often been ad hoc. The content of the course is designed to make you aware of the issues involved in selecting a tool and attempts to formalise the selection process.
There are five main components to the course:
There is now a large variety of KBS tools on the market. It would be difficult, if not impossible, for one to get to know them all before making a selection. This component identifies key tool features, so that when one faces a tool selection problem only certain category of tools will need to be looked at in detail.
In order to select a suitable tool it is necessary to consider the application to which the tool will be applied. This component investigates ways of analysing an application and examines the main application characteristics that affect the choice of tools.
This component looks at the mapping between application characteristics and tool features. It also highlights some of the non-AI factors - hardware, software and non-technical factors - that affect the choice of tools.
To give an overall flavour of the tools that are currently available, this final component reviews some representative tools belonging to different categories. In addition to the knowledge representation and inference facilities offered by these tools, their ability to interface with conventional languages and packages are also considered.
The course is scheduled to run on the 31 May. It is provided free of charge to academic engineers. If you would like to be registered on it or would like information about other services provided under the AI Support for Engineers project please contact Mrs Karen McGregor
The Latex documentation preparation system is now widely used in the academic community. Latex is a version of the TeX typesetting program which has been adapted for easier use by the layman, and has the advantage of being public domain. It is particularly popular with mathematicians and scientific researchers because of its ability to set mathematical formulae.
Enquiries have been received about a beginners' course and some thought has been given to the matter. A possible program would include:
If required, emphasis could be given to certain aspects such as maths or tables. Prepared handouts, giving examples of the more complex items, would be provided. Ideally, a one-day course would be conducted "hands on" at workstations equipped with previewers so that students could check their output immediately.
Readers who are interested in attending such a course, or in organising one at their local site, should contact me.
The Programmer's Hierarchical Interactive Graphics System (PHIGS) was published as an ISO/IEC standard for computer graphics in 1989. PHIGS attempts to combine a modelling system and a viewing system in a single standard. Implementation of PHIGS are appearing on the market, particularly on workstations (Sun, DEC, IBM, HP etc).
This tutorial, spread over eight hours of lectures, gives an introduction to the ISO/IEC standards for computer graphics, will cover the essential features of PHIGS, and will conclude with a review of the limitations of PRIGS and the steps taken by ISO to extend PHIGS. The final lecture will also cover the status of other projects in ISO.
The tutorial is aimed at applications programmes and users of graphics systems. Some familiarity with GKS would be an advantage but is not essential.
Attendance at this event will be free and assistance in locating overnight accommodation will be available on request.
As part of the continuing work of the Computer Algebra Support Project, the University of Liverpool Computer Laboratory is organising two one-day courses on computer algebra. Each course will consist of two talks plus a hands-on practical session.
The schedule for each course is as follows:
There is no charge for attending these courses, but since places for the practical session will be limited, advance booking is necessary.
