There were a number of changes in Group membership during this quarter. Frankie Sims left to take up an appointment with ICL at Bracknell and Penny Tucker, of ICL, finished her attachment to the Group. The Graphics Section increased its complement by the addition of Andy Dennis at the end of November. Efforts to recruit staff were not very successful and another advertisement will go out sometime in January.
Present group structure is:-
Most of the quarter was spent in the final debugging of the 7020 emulator. The problem of incorrectly formatted lineprinter output was finally cured. Hardware problems on both the RGO and Southampton 2050s hindered progress. The cassette recorder on the Southampton machine has had hardware faults, while the RGO machine had problems with the paper low condition on the lineprinter. Bugs were also obscured by differences in the configuration at ACL and the other sites which meant that the emulator worked correctly from ACL and Borehamwood, but not from the other sites. The major bug finally turned out to be a GEORGE error. The 7903 transmits a record of 81 data characters during the listing of a workfile, whereas the maximum should be 80.
RGO are currently using the 7020 emulator successfully. However, Southampton, with an identical configuration, have been unable to make it work fully. MDF is looking into the problem which appears to be either hardware or line trouble. FVS has written a more sophisticated line test program which should be useful in sorting out Southampton's problems.
Documentation of most of the emulator was finished by FVS. The main work during the next quarter will be to sort out Southampton's hardware problems and to produce a cleaner and more robust version of the emulator.
The full graphics capability of the Tektronix storage tube has been accessed using conceptual multiplexors on the 1906A. It is hoped that a similar facility will be allowed via the GEORGE LISTFILE system. If successful, the Tektronix will be added as another graphical output device for the SPROGS system.
A significant part of the quarter was spent in assessing machines for the Front End Processor. The aim is to connect the intelligent RJE stations to a Front End Processor which is, itself, connected both to the 1906A and 360/195. This will allow users to access both machines from a single RJE station. The main contenders for the project were the ICL 7905, CTL Modular One and the GEC 4080. A paper requesting permission to purchase a GEC 4080 was accepted by the Atlas Computer Committee. There are still some contractual problems to sort out but it is hoped that an order will be placed during this quarter with a possible delivery in the summer.
The hardware interface to the 1906A will be built in the Laboratory. The design will be similar to an interface built at Oxford University to connect its 1906A to a Modular One. Some problems may arise in obtaining the necessary interface cable to the 1906A and the special circuits to terminate the cable.
Most of the next quarter will be spent in understanding the 4080 hardware and software. A detailed study of both the 1906A communications software and the HASP RJE software will need to be made in the next quarter.
A number of discussions took place with the Rutherford Laboratory to consider the long-term aims of the communications work of both groups and to see how these fitted into any national networking scheme that might arise in the next few years.
During the last 3 months, a great deal of effort has been expended in keeping GEORGE up to date. Two new marks, 7.6 and 7.7 have been introduced. Both marks required a considerable number of changes to ACL's own modifications where conflicts arose. Introduction of Mark 7.6 was delayed for several weeks while serious bugs in the released version were sorted out. Mark 7 is now firmly established and the evidence is that it is a significant improvement over Mark 6, both in facilities and reliability. The introduction of Mark 8.10 is unlikely before May 1974, which should mean that the next quarter will be reasonably quiet. The reliability of Mark 7, together with considerable engineering effort to remove a number of irritating hardware faults, has meant a considerable improvement in performance over the quarter. It is a pity that this has not been reflected in the performance figures. This is partly due to lack of work, but mainly due to the disrupting effect of the UK4 processing.
The main software release during the last quarter has been the TASK system, which provides users with a standard compiler/consolidator interface for most of the main languages. This has been used internally for about two months and has been very well received. Once it is released to external users, it should be easier for them to get started on the 1906A.
A number of ACL's GEORGE enhancements have been taken up by other sites. In particular, there has been close liaison between ACL and ICL Bracknell. This has proved quite valuable and it is hoped that ACL's work has influenced the future plans for GEORGE. In this connection, a valuable meeting was held between ICL and ACL to exchange ideas on the 1906A, indicating where improvements were desirable in both the hardware and software.
There have been a number of small changes during the quarter. The automatic allocation of directories to small residences has proved quite successful. It has allowed a much closer control of channel loading. The split residence mends described in the last report have now been completed, together with the necessary changes to the Post Mortem routines. The main effect on the system has been a good MOP response time even in periods of heavy load.
The message EDIT OK is now sent to the monitoring file at the end of a successful GEORGE edit. This is for the benefit of the CORRECT macro.
The JOBTIME command has been enhanced to allow users to find out how much time they have used.
The input of binary cards into a file has been introduced. This is an off-line facility and should remove the need to on-line a card reader when inputting large amounts of survey data.
A number of modifications have been obtained from other sites (in particular, Manchester Regional Centre and Cardiff's CBC). These tend to be designed for GEORGE 3 systems so that there is usually a considerable effort needed to include them. Also, the originating site and ACL are often on different Marks of the operating system. Banner headlines have been added to ease the splitting up of output. Several other changes will be introduced in the next quarter.
The binary of the GEORGE message chapter has been analysed to produce a list of all GEORGE messages with their universal identifiers and message segment names. In particular, the subsets of SYSTEM JOURNAL messages and operator console messages were produced separately. In consultation with Operations Branch, it has been decided to remove or change a number of messages. The necessary mends have been produced but are not yet incorporated.
SYSTEM JOURNAL files have been building up in the filestore since the introduction of Mark 7. Mike Brown (Operations Branch) has implemented a system to COPYOUT these files to magnetic tape and maintain a date/time index to them. CJP has defined an interface to enable programs to access SYSTEM JOURNAL messages (whether files have been archived or not). This is being implemented both in ALGOL 68R and FORTRAN. Coding of the former is substantially complete.
The weekly user/job statistics program has been improved to give more information and to use the standard SYSTEM JOURNAL interface described in Section 3.4. This is run regularly to give the preceding week's statistics, but it can now be used to analyse any period since archiving of SYSTEM JOURNAL files began (1 September 1973).
An analysis of the distribution of size among listfiles was made from SYSTEM JOURNAL information. This indicated that a large proportion of the lineprinter output was produced in large documents and that monitor files did not produce a significant amount of output.
Enhancements to the GERONIMO system during this quarter have been confined to small alterations in facilities that were already present. The RETRIEVE command now supplies the tape serial number of the tape required for each file to be retrieved. This is useful to the operators in organising tape mounts when particular jobs are held up. The WLIST command specifies the PROPERTY qualifier of each file waiting to be listed at a remote printer.
A Console Operator Instruction has been issued. This provides all the information required to run the GERONIMO system and to cope with any error conditions that may arise.
The system is being prepared for delivery to ICL who will distribute it to sites which have requested it. Further documentation of the system is being produced as part of this exercise.
The manual was sent to the printers in December. Comments and suggestions are being received from those people who have draft copies. Once the manual has been issued for a period, it is likely that these comments will be collected and issued as an update to the original document.
A set of semicompiled routines has been created in the file :DYS.SEMICOMPJLIB, and documented in :DYD.JLIB-DOC. The routines include basic procedures to read or print a line, perform decimal conversions and handle recursion. The aim is to produce a set of routines which will reduce the labour involved in writing complete PLASYD systems.
Most of the work done on the consolidators so far, has been directed towards discovering the errors in the code supplied by ICL and working out suitable alterations. This task has not been eased by the total lack of documentation on the workings of the consolidators. Both XPCH and XPCK are now believed to be in a usable state, and work is being done to analyse the source code in order to discover what it does and how. As soon as some familiarity is gained with the structure of XPCH, an attempt will be made to insert code to deal with programs consisting of a number of overlays. This code is already available in XPCK and it will be used with suitable modifications.
A large amount of additional documentation was obtained from ICL on the internal workings of the compiler. Apart from a few minor problems, it has not been too difficult to produce a working compiler from the source supplied by ICL. Some work has been done in modifying the compiler.
The development of TASK has continued during the last quarter. It can now compile, consolidate and run binaries for both normal and overlay programs in ALGOL, FORTRAN, PLAN and PLASYD. Both the GROATS and SPROGS systems use TASK from within their own macros. It is intended that both GROATS and MVSL will be added to the set of compilers available from TASK. Some work has been done towards incorporating ALGOL 68R. However, the non-standard environment of this compiler tends to make it incompatible with that provided by TASK.
The TASK manual has been printed and is being issued. The main FORTRAN macro has been replaced by TASK. Currently, about 1,000 FORTRAN jobs a week are being run using the TASK system.
Several facilities have been enhanced in the light of user experience and a number of additions have been made. These are mostly to trap obscure failings of various systems and give meaningful diagnostics. (The TASK system itself has proved to be very reliable.) Minor enhancements include a PMD system for ALGOL which has the unusual feature of recognising a program requiring PMD without the user specifying this directly to TASK. A special facility to load several binary programs into a virtual address space has been implemented. This facility has been used by JDT to produce a utility which compares the code of two programs in binary form and indicates the differences.
A major enhancement has been to the peripheral assigning #MT parameter which allows a user to on-line actual magnetic tapes of any nature to a program using a much easier format than that provided by GEORGE. It will also obtain new magnetic tapes for a user. The peripheral assignment system will also recognise all assignment qualifiers and is, therefore, able to provide a common format for both ASSIGN and ONLINE commands, the latter being distinguished by parentheses around the name of the exofile or magnetic tape. Hence, for example, the restriction on qualifiers for exofiles and magnetic tapes for writing is now removed.
TASK will now accept parameters enclosed within parentheses (for one or more tasks), and this eases the limitation on the number of parameters the macro will accept and aids the interfacing of other systems on to TASK. It is planned to raise the parameter limit to 24 by use of this facility within the TASK macro.
Comprehensive timing and statistics collection has also been added. Analysis of this data, together with evaluation of the use of exofiles for the TASK compilers and their semi-compiled output is planned for the next quarter.
The original system for running jobs under :MANAGER has been extensively re-written to increase its security and enable a change to :MANAGER's password to be made with the minimum of inconvenience. Facilities now exist for certain users to run JUGGERNAUT, ACCOUNTS, LSTR and GERONIMO without any knowledge of :MANAGER's password. Other budget-controlling facilities are to be added to this system in the next quarter. It is also planned to extend the system to other high security users.
The New UTilities System, NUTS, is being developed. A paper describing the possible implementation and some of the problems of a comprehensive utilities system has been produced and feedback is currently being received about it.
Based on the experience with TASK, it was decided to plan the system from the outset to be capable of expansion to a significant size and to make it easy to use and simple to include new facilities. NUTS will consist of a kernel which will provide all the basic facilities (peripheral assignment, parameter reading, buffer space, etc), and each utility will interface onto this. It should provide a useful environment for the utility writer and ensure that the user interface to the utilities will be substantially the same.
The initial design has already been completed. Implementation is likely to start in the next quarter.
The Leeds benchmark was run on the 1906A for Glasgow University in order to obtain listings of the complete set of jobs run and their results (in the past, these have been suppressed to save paper). Comparisons between these Mark 7 results and earlier Mark 6 results showed that the Mark 7 ones were faster. Unfortunately, high error rates on the MOP simulator, when running under Mark 7, made it difficult to get a time comparison. The reasons for the errors are unknown and it was not felt worthwhile investigating them.
A new, larger version of the library update program has been received from Nottingham. Although this version is still too small for SPROGS and the other large systems, it is hoped that this version will be sufficient for most users. The new version corrects a bug in reading PLAN 4 semi-compiled segments.
This version is the first one that will be made generally available to users. A new SPROGS manual has been written and is currently in the final stages of proof reading. It has been decided that a glossy manual should be produced with a wide distribution. Most of the initial design ideas have now been incorporated. The main additions in the last quarter are:
Considerable effort has been expended during the quarter in trying to minimise the amount of store used. A number of subroutines have been converted to functions, thus saving an argument. The two largest alterations have been the removal of an initial filestore. allocation of 15K (the system now expands as required), and the use of COMPRESS INTEGER mode. The latter has meant changing all text handling routine parameters to REAL and altering the PLAN routines which access integer arrays.
The spooling system for sending SD4020 output from small graphics jobs to disc has now been implemented on both GROATS and SPROGS. It makes both systems easier to run for small jobs and should result in a faster turnround and much less loading and unloading of tapes. The spooled output is eventually written to one of a set of tapes owned by the SD4020 operators.
In order to stop people removing worktapes from the system, large amounts of output will now be generated on user-owned tapes. This has caused some problems for the SPROGS system which uses the TASK system to on-line the magnetic tape. It has been necessary for SPROGS to rewind the tape and re-read the title block in order to find out the tape number.
The LOWER data requirements of the routines for spooling in SPROGS have been considerably reduced by constructing the PERIs for on-lining the EDS file or magnetic tape in line rather than storing them in LOWER. These routines are only entered once so that the time overhead is small but the space saving is significant.
Using TREE-META, a preprocessor has been written and included in the SPROGS macro. It is a simple system which cuts down the amount of typing required in a SPROGS program and makes the program much easier to understand. Since it allows normal FORTRAN statements to be copied without alteration, it could be applicable to other programs. The overhead appears to be quite small. Its main features are:
File names and index variable names may be used explicitly without recourse to functions ANAME, PR and AR. For example:
NUMBR(1.0) CALL DRAW (ANAME (;FRED' ))
may be written:
DRAW ( 'FRED' /1)
Similarly, the statement:
CALL VEC(1.O,2.0,PR3(ANAME('VECARG')),4.0)
may be written:
VEC(1,2,£VECARG,4)
These simple modifications make film definitions in SPROGS much more readable.
Two Hewlett-Packard teletype-compatible pen plotters have been ordered as a means by which remote users can get small amounts of graphical output immediately. The SPROGS system has been modified to allow these as additional output devices. Testing of this facility will proceed as soon as the plotters arrive.
Progress on the film has, to some extent, been curtailed by preparations to issue Version II. However, film sequences depicting the SPROGS file system and sequence list have been produced. The method adopted has been to define individual film and picture files needed in a sequence initially and then to provide a controlling film file (known as the steering file) to generate the output. Sequence lists are used to obtain any parallel processing required. It is evident that the system does have the ability to split up the film definition into its constituent parts of object definition, movement definition, timing and control allowing the user to concentrate on each aspect in turn.
The current system does not contain some of the basic animation facilities such as wipes and holds. Consequently, these are being developed as specific routines for the SPROGS film. They will be added to the basic system as display routines at a later date. Generating holds proved to be quite simple. However, the facility of saving actual SD4020 orders has proved quite important. In the sequences generated so far, the saving in machine time using this facility seems to be about 40 to 50%. The main film sequences are currently being linked by a right-left wipe.
Some simple fairing algorithms have been produced for handling start and stop motion so that it does not appear jerky. These are currently implemented as film files, which are drawn from within the steering file. More sophisticated algorithms will be added at a later date.
Hardware faults are still a major concern on the PDP15. This has meant that for yet another quarter, programming staff have been spending more time being involved with the engineers than writing software. To try to alleviate this situation, it has been agreed with Operations Branch that the day-to-day running and maintenance of the PDP15 is to be gradually taken over by them. At the moment, Alan Hancock is being educated by JRG and WDS in the use of the machine. He is currently producing some display routines. One of these should emerge as a picture-driven program to exercise the whole machine and its peripherals. This is quite an appropriate time for this transfer of control to take place. The additional core and disc store has been installed and the new RSX-3 operating system will soon be available. The machine should then be in a reasonably stable state for some time.
During this quarter an extra 32K of core store was added to make a total of 64K words. The new store is slower and much cheaper than the original. The cycle time for-the new store is 980 nanoseconds, compared with 800 nanoseconds for the old one. The combined effect of the slower core speed, the memory multiplexor necessary to attach memory above 32K, and memory relocation and protection, is to make the cycle time about 1.16 microseconds. The store above 32K cannot be addressed indirectly by the CPU. It can be addressed by using an index register, by relocating, or by addressing indirectly using the VT15. The extra store will be used to accommodate more than one user task under RSX-3.
Two more fixed-head disc surfaces were added to make a total of four surfaces (1 million 18-bit words). About 1-1.5 surfaces will hold the operating system, RSX-3, and the software associated with it. Most of the remainder will, eventually, be used as a large, fast scratch area.
An EIA converter (LTI9C) has been added to one of the two teletype ports of the LT19 multiplexor. This will allow an EIA-compatib1e Tektronix storage tube to be attached. The interface has been tested using an ACL-supplied teletype. Unfortunately, in order that the PDP15 will work, this teletype has to be switched on even if it is not being used, otherwise unexpected interrupts are generated which DOS cannot handle.
An ECO (Engineering Change Order from DEC) has been added to the VT15 to enable it to address 17 bits indirectly. The VT15 did not need this facility before the extra store was added. It has meant that minor changes have had to be made to the basic graphics software which, in turn, revealed a long-standing but dormant bug in the tape viewing program.
Only a few new hardware faults occurred during this quarter and most of these were simple component failures. The outstanding problems at the moment are:
The first four of these problems remain from the last quarter. The problems are known to DEC and are receiving some kind of attention. During this quarter, DEC were asked if they would perform bi-monthly preventative maintenance on the machine and set up a schedule for performing ECOs as they were received. DEC did not seem to think that regular preventative maintenance would have located or prevented many of the faults that occurred. They also stated that all ECOs had been performed on the machine. However, in December 1973, several VT15 ECOs were installed. One of them was released by DEC USA in September 1972!
RSX-3 was expected just before Christmas but the DECtapes sent from Maynard were held up in Customs (Resource Smuggling Executive ?). The basic graphics routines, FOG, have been split into smaller files. The system is slightly larger, but it is now possible to leave the bulk of FOG in an overlay segment if core space is short. It will necessitate the program being written in a particular way.
Routines for retrieving lists of lightbuttons (menus) from the PDP15 disc are working. A preliminary version of PIGS including automatic menuing is available as a demonstration program. The first release of the system (under DOS) will be available when the assembler for the PIGS Menu Definition Language (MDL) is completed.
The panel for the interface unit has now been built and installed by the Rutherford Laboratory. The interface unit has had power applied to it and appears to have no short circuits. Some wiring faults have been discovered. NMP has obtained wire wrapping and unwrapping tools and corrected these faults. Now that the new PDP15 cabinets have arrived, the interface unit will be tested on the PDP15 in the near future.
The AUTOPM problem with the Mark 7 Executive (reported in the last report) was finally resolved at the end of November. This meant that two months were lost while this one software fault was cured. This delay is partly due to the time it takes for an Executive fault to be reported to ICL Bracknell - posted to West Gorton - posted back returned to site. The other and major problem is that no testing of Executive mends is made at West Gorton; as a similar device is not available. Consequently, several proposed mends did not cure the bug. At some stage in the process, some on-site testing and mending would have speeded things up. With Executive writers moving to Bracknell in the near future, perhaps less delay will be experienced in the future.
The first version of the software is now capable of sending character files from the PDP15 to the 1906A. The system is being checked out to ensure that all programs in the PDP15 which generate characters can use the BSI.
The next version of the software will read binary information from the 1906A and, as an application, the tape viewing program will be enhanced to read from the graphic spooling file on the 1906A. Once that is working, the full system for sending both characters and binary information in both directions will be implemented.
It has been decided, after all, to hold the COLAB training games at ACL in 1974. As the courses are taking place at High Wycombe, it does mean transporting the people to and from the Laboratory. It has also been decided to allow only half a day for the game (this may prove to be too short). The new arrangement does mean, however, that the course students will be able to play with teletypes.
A number of experiments in the use of conceptual multiplexors have been carried out with a view to using one to control the teletypes for COLAB. There are still problems with conceptual multiplexors and these experiments produced a number of GEORGE breaks. As long as the program has only one member, it appears as though such devices work correctly but slowly. Multi-member programs, however, proceed very slowly indeed and either produce a GEORGE break or hang. up inside GEORGE on a PERI order to the inter-processor buffer. When one hangs up, it is very difficult to abandon the job.
A paper describing COLAB and its use, written by RET and DCT, has been accepted for publication in Software - Practice and Experience.
Visit to USA, C J Pavelin
The Group held another successful Skittles evening at the Hare and Hounds, Chievely, on 7 December.