Front cover shows a 3D electron density surface of a phenyl alamine dipeptide molecule, sliced through at the top to show part of the side chains. Coordinates provided by Julia Goodfellow of Birkbeck College.
In recent years I have written year-end messages which have appeared in the November/December issues of FLAGSHIP. This year, for no great reasons of principle, the message is appearing in the new year, so let me start by wishing all readers of FLAGSHIP a very happy and prosperous new year.
1993 seems to have been at least as busy as any other I can remember, and in some areas of our computing services we have been shifting ground fairly rapidly. Last year I noted that we were starting to introduce new superscalar facilities around our IBM 3090 that would be based more on the RISC technology price/performance curve. As a number of articles in recent issues have shown, these facilities have come on well during the year and in aggregate they now provide substantially more scalar computing capacity than the 3090 ever did. The first of the new facilities, the Open VMS service, is proving robust and has developed quite a broad user base. The more recent Unix-based facilities, the OSF/l cluster and the CSF farm, have been running in a pre-production mode for some weeks and will soon become full production services. The scalability of all these facilities at economic costs is very attractive, and their modularity can enable adjustments to capacity to be made in accordance with demand at much shorter notice than was possible with classic mainframe technology. In a similar vein, it is interesting to see that developments in storage technology are also now opening up new options for dealing with very large quantities of data, and we are looking into ways in which they might be exploited to meet the growing needs. At present we have around 10 Tbytes of data in the building here, of which about 4 Tbytes is on-line. The amount of data stored is growing by about 25% per year, so the attractions of any aid that new technologies can offer in helping cope in this situation are clear.
On the supercomputing side of our activities, the past year has been one of reasonable stability. The Cray Y-MP has delivered a solid service. It is loaded just about up to its capacity and the forward demand from peer-reviewed projects now exceeds capacity. Taking the national facilities as a whole, some relief on the demand for resource should be felt from the bringing into service of the recently installed Fujitsu supercomputer at Manchester, and of course substantial further capacity will be injected into the system when the current procurement of new national high performance computing facilities has been completed.
A major development in 1993 was the arrival of SuperJANET, which opens up opportunities for more bandwidth-intensive modes of working between central supercomputers and workstations or other machines local to the user. In order to gain some experience of what is possible, a small number of Cray users took part in SuperJANET pilot projects in which pictorial information computed at the Cray was worked with interactively from workstations in Cambridge and in various departments in London. The pilots were successful in that they worked, they demonstrated the sorts of thing that could be done, and they gave a feel for the amount of effort that would be required to make this kind of working routine.
Turning away from the immediate technical issues, the Government's 1993 White Paper Realising our Potential - a Strategy for Science, Engineering and Technology has set in train some rather fundamental changes to the structure and missions of Research Councils which will start to take effect in 1994. As I write this the new structures are not in place and the detailed consequences of the changes are yet to be made known. The one thing that probably is clear, however, is that the rate of change is unlikely to lessen in 1994.
In times gone by, when supercomputers were unsocial creatures and networks were slow and unreliable, the only way to use supercomputers was by submitting batch jobs from front end machines. Fortunately for all of us there has been a lot of progress in making supercomputers easier to use and it is time to have a serious look at how much we really need these front end services.
Since the introduction of Unix-based operating systems, TCP/IP wide area networks, and windowing PCs and workstations on most desks, it is now very easy for all our Cray users to log in to Unicos and do all their editing, housekeeping and development work from an interactive session. It can be argued that using a supercomputer to edit files is a waste of machine cycles, but the overhead of interactive sessions is small, particularly since we have added the 256 MWord Solid-state Storage Device (SSD), and in most cases the gain in end-user productivity justifies the small cost of the CPU time.
Our aim is to make the Atlas supercomputer services as easy to use as a local workstation but with the power and data capacity of a major supercomputer. The Unicos service is very reliable and we are seeking to reduce the periods of planned maintenance and system development even further. We would like to encourage all our users to try using Unicos interactively and to find out that in most ways it is just like a big workstation. You can compile and debug interactively and run jobs up to 600 seconds and 6 MWords from the command prompt. If you find these limits restrictive please let us know and we will see what is possible in individual cases.
For the historical reasons above, particularly true when we were running the Cray X-MP/416, we have three different front end machines in use (Unix/RS6000, V AX/VMS and IBM VM/CMS). Each provides a distinctive service but we have to ask for how long these services are really needed.
Even though our front end machines have all been paid for, they consume money in maintenance and software licences, and human effort in fixing problems and backing up filestores. On each service the balance of costs and benefits is different:
The Unix/RS6000 front end (unixfe) is host to a number of other services used on and off the RAL site; being a modestly powerful workstation its maintenance cost is not too high and it offers a platform for graphics and other Unix software which is not always available for Unicos. The manpower costs of running unixfe have come down as the service has matured.
The VAX/VMS front end (vmsfe) has been trimmed over the last year to remove the very expensive and not very powerful VAX 8250 supercomputer gateway. The remaining VAX 3600 is still expensive to run in terms of maintenance since it is an old technology machine, but it consumes minimal manpower.
The IBM 3090 running VM/CMS provides many other essential services for SERC's corporate and scientific computing. As explained in recent issues of FLAGSHIP, we are introducing new facilities for handling the scientific workload that has been running on the IBM and we are now looking at what we should do with the other functions that the IBM supports. One of these is RQS/VM, the Cray job submission software, and we need to know whether it can be phased out.
As well as the local front end machines, we have support costs in money and manpower for remote job submission using both JTMP and FTP over the X.25 coloured book protocols and Cray's modified ftp process over TCP/IP.
In the long term there is very little need for front end services for any but the most exotic of supercomputers, and as Unix is widely accepted as the basis for most scientific computing we would expect to see the situation where a general-purpose Unix server would be the only front end machine that is needed. In the past we needed front end machines to service the user community that only had access to X.25 Coloured Book software for networking but today everyone appears to have access to TCP/IP software in the shape of telnet, ftp and, in many cases, xterm.
The requirement now is to establish the time scales over which other services can be phased out. As reported in a previous issue of FLAGSHIP, the software support available for Coloured Book protocols on many machines is rapidly disappearing and we will move into a situation where new services at the Atlas Centre may only support TCP/IP access.
There are no changes imminent, but in view of the decreasing fraction of work that is submitted from front end machines, we are soliciting the views of the user community: through this article, through news files, and through the Cray User Meeting in Spring 1994. The earliest date at which existing front end services could be phased out would be towards the end of 1994.
If the VM and VMS front end services were withdrawn we would lose the existing X.25 access to the supercomputer services, but in case there is anyone who is not connected to the JANET IP service, we will maintain an X.25 to telnet gateway machine for interactive sessions (see FLAGSHIP 28).
We need to strike the right balance between providing what people want and making good use of our budget; obviously, the more people want and need a particular service, the more resources we can justify in its provision. Please write or e-mail your comments on the usefulness of the Cray front end machines and join in the debate at the next Cray User Meeting.
The DEC Alpha Farm running OSF/1 will become a production service in mid December. The configuration is still as described in FLAGSHIP 28, but extra memory has been ordered for one of the back-end machines. By increasing the memory to 256 Megabytes we hope to make this farm suitable for running larger memory jobs.
The hardware consists of five DEC 3000 model 400 workstations connected together by FDDI and also connected into the RAL site FDDI network.
In addition to the software listed in the previous issue of FLAGSHIP, the following have been installed:
Version 3.2 of the Parallel Virtual Machine (PVM) message passing software from the University of Tennessee and Oak Ridge National Laboratory has been installed. PVM allows a single application to run across a number of physical machines, in effect turning them into a parallel computer. It is implemented as Fortran-and C-callable libraries which allow a user program to launch previously-compiled binaries on remote systems and to communicate with them and pass data.
PVM programs can run on a collection of heterogeneous systems so the farm could interwork with machines at your institute. However, network bandwidth is then an important factor governing the overall performance, so unless your PVM program is carefully designed to send a minimum of data across the network, we would not advise its use in this way.
The Atlas Centre OSF/1 Alpha Farm processors are interconnected by FDDI and the latency between two machines has been measured at 2.3 milliseconds with a bandwidth for large transfers of 1.8 Megabytes/second. This latter figure is well below the theoretical limit for FDDI. This is probably due to the overhead of PVM message passing but some further tuning of FDDI may be possible.
The main focus of this Alpha Farm is to provide a multi-user service, so we cannot guarantee to give dedicated use of several CPUs to a single user. However, the farm is well-suited to the development of PVM programs.
All existing users of Atlas Centre services are eligible to have a userid on this service. SERC grants can be transferred from other services and new grant applications can be made to the February grant round. Users funded by other Research Councils should contact me for advice. Pump-priming applications should be sent Dr B W Davies at the Atlas Centre, other enquiries about the service to John.Gordon@rl.ac.uk.
The SuperJANET backbone network has now been operational for about six months, and the range of applications continues to grow. Many of these are scientific applications, but there are also the medical ones. Historically, the Royal Postgraduate Medical School (RPMS) was the first institution to be connected to the backbone network (see Figure 1) and has been closely associated with the development of high bandwidth medical applications. A one-day conference on the Medical Applications of Super]ANET was held on November 9th at Hammersmith Hospital.
Our intention was to hold not a technical meeting describing how things could be done, but rather a series of demonstrations showing what could be done over SuperJANET. Because we wanted potential medical users we only advertised the event in the medical press. In the event, the audience consisted of about a hundred doctors from the UK and abroad, and some senior hospital managers. There were also a similar number of other interested parties from a variety of organisations.
One problem with trying to demonstrate the range of applications which can be run on a national network, is that you need to provide a range of platforms for running them on, since all users, almost by definition, choose different hardware. Given the scale of the meeting it was perhaps unsurprising that we ended up with a PC, a powerful Macintosh system including video disk player, two Silicon Graphics workstations and three Sun workstations, to say nothing of a laser printer, document scanner and microscope with its remote controller and associated TV. A large bench was required to accommodate the kit. Cabling it also required some thought. Finally, the projection of a multiplicity of screens so that they can be viewed by a large audience is an expensive matter: we used three video projectors in total.
The major technical problems to be solved concerned piping video from the SuperJANET network to Hammersmith. For various arcane technical reasons the only practicable solution in the time available turned out to be sending 2 Mbps compressed video from UCL to Imperial College (over a Megastream put in for the occasion by BT) and uncompressing it there. The analogue signal was then turned into an A TM cell stream using a Spare station fitted with a Parallax card and transmitted through a Fore Systems A TM switch onto WestNET (see Figure 2). This solution was technically inelegant, but at least it worked.
In fact, the only significant problem experienced during the meeting was at the low tech end of the spectrum - a humble PC, which had performed impeccably in rehearsal at lunchtime, refused to do its stuff when called on during a demonstration about an hour later. Such is life.
Cytovision is an interactive, multimedia training package which is aimed at improving standardisation in the National Cervical Screening Programme. The package contains a comprehensive library of microscope images, together with video sequences and interactive animations, which provide detailed coverage of all elements of the screening programme. Lesley Turnbull and Roy Stringer (Liverpool) gave a demonstration of some of the innovative features of Cytovision, which has now matured from a research project to a commercial package.
Henry Rzepa (Imperial College) gave his usual polished presentation of the potential for multimedia to assist in molecular research. He also showed how two or more collaborating sites can discuss molecular information, using a combination of videoconferencing and real time molecular editors and visualisers. Finally, he demonstrated how scientific work could be converted to animated graphics and archived on gopher servers for rapid global access.
It was interesting to note that both multimedia demonstrations had been developed on the Macintosh, which must say something about its abilities in this field.
Modern drug design depends not only on the bench chemist but also on using computers to model the shape and chemical characteristics of potential drug molecules and their targets. Julia Goodfellow (Birkbeck) showed examples of the use of the UNICHEM package to perform real time quantum mechanical calculations on the Cray Y-MP at RAL. Although most of the algorithms produce vast quantities of output, it is usually more helpful to visualize the molecular shape or properties on a graphics workstation. In the future SuperJANET will facilitate the visualization of even larger changes in structure and realtime interactions.
Hyperthermia is a promising adjunct to radiotherapy or chemotherapy in the treatment of some cancers. Heating is usually produced by microwaves or ultrasound, the advantage of the latter being much deeper penetration into tissue. The development of ultrasound heating equipment and of treatment planning methods are both active areas of research at Hammersmith. Supercomputers are necessary given the scale of the calculations.
Jeff Hand (RPMS) showed patterns of power deposition from a device designed to heat a tumour in the prostate. The calculations were performed in real time on the CONVEX 3840 supercomputer at ULCC. Preliminary results obtained with the device in a 3-D model of the human pelvis (derived from serial CT scans) were displayed using the A VS software package. Without supercomputer access the feasibility of this approach could not have been proved, an essential step before construction, and eventually clinical trials, can begin.
For both projects SuperJANET provided the link to harness the power of the national supercomputers to the workstation on the research worker's desk.
Much of medical practice has to do with the interpretation of visual images and this is especially true in the field of histopathology. In some circumstances a local pathologist may not be able to reach a firm diagnosis, or would like a second opinion, which using conventional methods (e.g. the postal system) can take considerable time. The PathNET project aims to make pathology expertise more rapidly available by network transfer of high quality images. Pete Mills (Manchester) showed a remote consultation between Prof McClure in Manchester and a local pathologist at Hammersmith, both of whom could view the same microscope slide. They each had independent pointers for delineating features as well as a two way audio and video link for videoconferencing.
At the RPMS we have also developed a system for collecting and transmitting pathological images through the SuperJANET network. In addition ours provides a facility whereby the microscope from which the images originate can be controlled remotely. Chris Foster (RPMS) demonstrated a remote consultation with a pathologist in Edinburgh, Dr Edward Duvall, who was also able to manipulate the microscope in real time.
Combining the best features of both projects would provide an extremely powerful tool for telepathology.
TACT is a planning tool developed for brain surgery. Louis Lemieux (RPMS) showed the interactive registration of brain images from the same patient using 2-D and 3-D datasets from scanners at Hammersmith and Queen Square. Processing these large datasets was carried out in real time. An important advantage of fast networking is that big datasets can be manipulated without the necessity for copying them to the local workstation. An additional advantage is the possibility of replacing the surgical planning workstation, normally located in the operating theatre, with a cheaper and more robust device, such as an X-terminal.
Scanning with Positron Emission Tomography (PET) provides another means of producing 3-D images of the human brain. Such functional images become even more meaningful when combined with the structural images produced by magnetic resonance imaging (MRI). The resulting volume dataset is, however, large, often exceeding 6 MBytes, and full interpretation is only possible when it can be manipulated interactively. Ralph Myers (Hammersmith) showed an example of a 3-D dataset, acquired while activating the articulatory loop which is involved in short term memory.
Foetal medicine deals with the diagnosis and treatment of diseases in the unborn child. Because of the rarity and complexity of these problems there are only a handful of specialist foetal medicine centres in the UK. This creates two difficulties: first, pregnant women anxious about their babies often need to travel long distances for a specialist opinion and treatment. Second, difficult problems are dealt with by subspecialists, such as paediatric cardiologists or urologists. However, they often have to base their opinions on still pictures, rather than watching live ultrasound scans of the baby.
Sarah Bower (Queen Charlotte's) showed a live ultrasound scan of a foetus in utero at Queen Charlotte's Hospital, about 2.5 km from Hammersmith. The demonstration also showed how Dr Janet Vaughan in the scanning room could consult the remote expert at Hammersmith across the network.
In networked anaesthesia the network is used to monitor an anaesthetic, rather than to give one. A system for the remote monitoring of patients in the operating theatre has been developed at Hammersmith. Haemodynamic information from patients undergoing surgery is acquired every two seconds, stored in a computer, and processed to display trend information for the anaesthetist. This provides an accurate record of patient data and of the start and stop times of operations, which can subsequently be processed for audit purposes, e.g. to see how many episodes of significant hypotension occur during induction of anaesthesia.
David Harris (RPMS) showed a remote consultation, using both data and video links, between a senior anaesthetist in the auditorium and a junior anaesthetist in an operating theatre at Hammersmith Hospital. SuperJANET will allow the relaying of data to distant experts for improved supervision of trainees, and may assist the development of collaborative research within the UK.
UCL have had a lot of experience with the use of video teaching methods, having been involved with the London University liveNET project since its inception. The arrival of SuperJANET has offered them wider scope and for the last few months undergraduate audiences at Cambridge, Manchester and Edinburgh have been able to receive video from operating theatres at the Middlesex and to ask questions of the surgeons.
At Hammersmith, Gordon Jameson (UCL) and Prof Mike Hobsley (UCMSM) showed a live teaching session between the operating theatre at the Middlesex Hospital and a group of medical students at Edinburgh University.
The experience of putting on SuperJANET demonstrations for the JNT earlier in the year was extremely valuable and enabled us to plan a technically more ambitious meeting than had previously been attempted. It goes without saying that thanks are due to a very large number of people (see below).
SuperJANET represents a quantum leap in technology and novel uses are only likely to increase. This applies particularly to medical work, which is heavily image-based, and especially well placed to take advantage of the new capacity. It must be stressed however that SuperJANET is an academic and research network and is not provided for routine NHS use. When the applications get sufficiently stable that we want to run them as routine it will be necessary to solve various current problems. These include payment for NHS use, data security and confidentiality, and the medico-legal implications for the service provider. None are, I believe, insuperable and the future of telemedicine in the UK looks exciting indeed.
I would like to thank all those involved in the conference: the commercial organisations who loaned us equipment for the day, and numerous individuals (besides the speakers themselves) from Computer Centres all around the country.
