Group membership remained the same during the quarter apart from the arrival on 6 August of a sandwich student, A J Pitchers, from Portsmouth Polytechnic. A temporary addition to the Group is Miss Penny Tucker of ICL, who is expected to be at ACL for about 3 months in order to gain site experience.
Present group structure is:-
The 7020 emulator on the 2050 is now working and jobs have been run successfully to the 1906A from the 2050. The final testing and debugging is currently taking place. It is a pleasure to record the group's thanks to Olive Parker from GEC who has contributed greatly to the project.
The emulator has also been used at GEC, Borehamwood; this has provided useful experience in the remote use of the software.
The cassette recorder on the 2050 has proved a satisfactory mechanism for storing the emulator and for transferring it from one installation to another.
Currently FVS and MDF are trying to implement the emulator at RGO but to date, hardware difficulties with the line and the modems have inhibited progress.
Plans are evolving for using 128 or 96 character line printers, MOP terminals and displays on the 2050.
The use of the 1906A by Imperial College is now satisfactory and has proved the ability to use MOP at 600 bit/sec.
Experiments are proceeding with conceptual multiplexors with a view to using Tektronix devices on the 1906A.
Plans are being formed to front end the 1906A with a substantial communications processor. This will allow users to access the 1906A or 360/195 from terminals and also allow more exotic use of the equipment in terms of MOP terminals and displays.
There is no doubt that this quarter has been the most encouraging on the 1906A. Several of the major projects have been released to users or are about to be. In addition the long drawn out introduction of GEORGE 4 Mark 7 has been accomplished.
Near the end of the last quarter, Ron Prior left the laboratory. He made a number of significant contributions to the 1906A operating system. ICL have loaned Miss Penny Tucker for a few weeks to rewrite the JUGGERNAUT retrieval system. This collaboration between ICL and ACL at the working level has proved quite valuable and it is hoped that similar arrangements can be made in the future. Similarly, closer liaison between the groups and Operations Branch has proved extremely useful.
The major event has been the introduction of GEORGE 4 Mark 7. The initial trials of the system were frustrating and showed that the Executive and GEORGE were subject to serious errors. Eventually, a plan of action was agreed (see last QPR) and ICL took several weekends testing Executive and GEORGE, an exercise which culminated in an invitation to use the ACL filestore in a test. Mercifully this was reasonably successful and Mark 7 was adopted. Although the error rate was worse than Mark 6 this did allow ACL to channel its meagre resources more fruitfully. A fairly rapid improvement followed as ACL removed the bugs in the private mends. Also, considerable effort was expended in finding a particular obscure ICL bug which led to random overwriting of programs.
This has been followed by a gradual improvement. There is now strong evidence to suggest that much of the poor performance has been due to hardware, in particular the tape system. A recent week, after extensive engineering on the units, yielded only 5 GEORGE systems breaks aga1nst a norm of some 20.
The addition of a 5-spind1e EDS 60 gave an opportunity to increase the amount of on-line GEORGE filestore (by 40%) and spread it over 2 disc controllers. Most of the EDS 30 are now permanently mounted, the non fi1estore-units being used for GEORGE chapters, libraries, and exofiles required by certain subsystems. Small residences, to contain directories, have been created on the outside few cylinders of each unit. A small residence has been set up on one of the drums for some of the GEORGE chapters (see below).
The efficiency under Mark 7 has been good.
Under Mark 6 the machine was disc limited. With the new directory structure under Mark 7 this was not so evident. The introduction of the EDS 60s also eased the channel limitation on the filestore.
Average object program efficiencies are now about 50% and further improvements are expected from splitting the GEORGE chapter residences. A reassessment of all installation parameter values is also now overdue; currently the amount of core for user programs is being increased.
The performance listing program was adjusted to cope with the increased backing store. The production of performance graphs has been automated and the results filed weekly.
Major changes are as follows:
A small filestore has been created in order to speed up the process of testing GEORGE mends.
It is hoped that few further major modifications to GEORGE will be necessary. Implementation of a Secure Job Well has been postponed since ICL and Oxford University are already pursuing independent efforts in this direction.
Further statistics have been collected on the amount of GEORGE workspace used for various purposes. . As this information does not appear to be significantly different under GEORGE Mark 7 from what it was under Mark 6, nothing further has been done. A program has been written to monitor GEORGE chapter usage, and to produce histograms indicating the average core residency and swapping rate of chapters in use. This has been used to determine which chapters to allocate to the drum in the split residence mends.
There has been a considerable increase in system security; in particular :MANAGER has been made a high security user. This means it is not possible to run work under that user other than from a MOP terminal (ie having quoted a password). For this reason a system has been developed to initiate programs which must be run under :MANAGER such as GERONIMO JUGGERNAUT and ACCOUNTS.
A program to produce weekly job statistics has been written and it is hoped to automate the process. The number of journal files began to get out of control but certain unimportant messages are now being removed and the target journal size has been increased.
Mike Brown (Operations Branch) has constructed an indexing system and is copying out raw journal files to magnetic tape. It looks as if the compacting problem will be less acute than feared since even now a magnetic tape will contain about a month's worth of files.
It is hoped that there will be time and effort available in the next quarter to define and implement the journal archiving scheme and improve the statistics production.
A scheme has been implemented in the HLS to issue engineers' test programs automatically in slack periods. Engineers can slot in an appropriate set of tests as required.
The method of determination of WAITFACTOR has been improved.
Several new facilities have been added to the GERONIMO system, and some of the more popular facilities have been enhanced. Those which have been enhanced are:
New commands introduced this quarter are:
There is also a facility for issuing operator commands from a GERONIMO terminal as if they came from the operators' console. Such commands are vetted in the interests of security. and. whenever possible, a reply is returned to the VDU.
Documentation of the TASK system was released to internal users at the start of the quarter and this resulted in several requests for enhancements. The largest such request was for the XPCK consolidator to be included so that overlay programs may be consolidated under TASK. This implementation has been extremely difficult resulting in a complete re-design of the interface between TASK and the compiler and consolidator binaries. The inclusion of this facility means that TASK also ought to be able to RUN such binaries. The binary program loader was thus re-written and will currently load normal programs written to direct access files by the consolidator (as opposed to saved binaries). The loading of overlay programs is now under investigation and is proving difficult.
Extensive work was also done to the LINK system to meet the requirements of large users using semi-compiled modules. The system now allows users to maintain all their programs in semi-compiled form, thus needing only to recompile the module containing any modifications and letting TASK LINK all the modules to produce a binary program.
The release of the documentation to selected external users resulted in the request for a property parameter (PR) so that lineprinter listings would be sent to the appropriate RJE station.
Many other parameters and features have been introduced to enhance facilities and to ease the interfacing of several other systems on to TASK. The various messages put out when errors occur have also been improved, with the aim of making failures self-explanatory. Feedback from users in this area is much appreciated.
Another major undertaking was the transfer of the TASK compilers to an exofile. Due to TASK using large PERI reads (up to 31K), this meant the compilers had to be specially laid out so that the relevant blocks were within a disc cylinder. This is now ready but introduction is being postponed until this, and the use of compilation and consolidation using exofiles for the semi-compiled, can be evaluated. Preliminary work in this area suggests that TASK reduces the GEORGE time for a FORTRAN job by more than 4 seconds.
Future work on TASK, in addition to items already mentioned, will be to improve the statistics logging and to write an analysis program for these. No further major enhancements to TASK are planned.
The latest release of the documentation (dated October 1973) is actually ahead of the implementation and is the basis for a user manual which is currently being printed.
NUTS is a scheme whereby all the frequently-used utilities will be grouped into a central system with a standard interface for each utility.
In this requirement it is very similar to TASK and is being developed from it. Hence all TASK developments, notably the rewriting of the binary program loader, is being done with this in mind.
Currently a feasibility study of such a combined system is being done and initial results suggest that there are many problems. Effort to implement such a system is unlikely to be available before the New Year.
This is a disc editing system designed to print the binary of GEORGE chapters, or make simple amendments to them, without the time and cost of a GIN compilation run. It can also be used to print or amend buckets of any I-block bucket disc file; in particular it provides a means of operating on the GEORGE filestore, if such a procedure is ever necessary to save a general restore.
It has been used extensively in testing large scale GEORGE mends.
The Nottingham library generation package was obtained to assist in the creation and maintenance of semi-compiled libraries. Tests have shown that the current version is somewhat restrictive, in that only 130 routines can be handled at once and no mutual recursion is allowed. A new version removing these restrictions should shortly be available from Nottingham.
Work on the source of XFEH obtained from Oxford was eventually abandoned as hopeless, and a fresh source has been obtained direct from ICL at Bracknell. Work is at present in progress on producing a working compiler, and, despite some problems, progress is being made with the assistance of Penny Tucker and ICL Bracknell.
Once it is certain that ICL have provided the source of a working. compiler, it will be possible to start amending the compiler as required.
The coding of the final parts of a new Juggernaut system to incorporate the required enhancements is being finished. Then the JUGGERNAUT programs will be linked together and the system testing carried out.
The FORTRAN analysis system has now been released to external users in a suitably modified form to make it consistent with ICL software. In response to user demand a routine to convert input from ATLAS or EBCDIC card code to standard 1900 code has been written and incorporated.
This has now been completed, and a draft version has been issued to internal users. It is proposed to allow time for comment and criticism before issuing the manual externally.
The TREE-META translator writing system has been implemented on the 1906A. The current system follows closely the design of the version available on the INTERDATA computer of Queen Mary College. It uses the same storage system for strings, trees and stacks. The method of storing and accessing strings is not very efficient and it is likely that this will have to be changed to speed up the final version.
A translator for a particular language L is defined in terms of two types of TREE-META statements. The first type defines the syntax of the language L and describes how trees will be built from sentences in the language at recognition time. The second type of statement describes how these trees will be scanned and manipulated to generate code equivalent to the input sentence.
So far the system is defined in terms of a set of basic routines written in PLASYD. Both types of TREE-META statements are compiled into calls of these basic routines so that the system is similar to an interpreter at present. To check-out the system, a definition of TREE-META in terms of itself has been successfully input to the system. A number of delays in the production were caused by internal PLASYD tables overflowing. In particular the limit of 100 on the number of unresolved forward references has required a large amount of code rearrangement.
The aim in the next quarter will be to speed up the system by redefinition so that PLASYD code is generated rather than routine calls. This is not too difficult as, although the system is large (about 6000 PLASYD statements), its definition in terms of itself is only about 150 lines. Once an efficient version is available, it will be used to experiment with preprocessor facilities for the SPROGS system.
This is a system, which originated from Leeds University, for assessing the effect of MOP usage on GEORGE by simulating up to 63 terminals on the 1906A. The system has been modified to run under MK7 and copies sent to Birmingham and Aston Universities. It has also been used to evaluate MOP interactive work and its effect on the machine and is to be run with the Leeds multi-access benchmark for Glasgow University.
Code has been produced to provide a list of all directory names and superiors in alphabetical order from MASTER DICTIONARY.
More people have begun to use this system in the last quarter with varying degrees of success. It appears to have been successfully incorporated into the BOON time series package but attempts to include it with ASCOP are running into difficulties over the size of lower data required. A version has been released to the Cambridge CAD centre for implementation under GEORGE 3.
Various new features have been added. It is now possible to rotate a whole region, and to correctly execute a DRAW from within a file definition. The format of file headers has been changed to provide a backward pointer to the appropriate file table entry, and a garbage collection system on the SPROGS filestore is under construction.
It is also possible to include basic device orders in a SPROGS file. This feature greatly speeds the generation of constant information from frame to frame by removing the need to perform repeated conversion and scissoring. A file block structure, using routines BEGINS, ENDS, has enabled a group of routines to be controlled by the conditional WHILE (see previous report), and has proved a considerable aid in the readability of files.
The problem of lower data has been combated to some extent by the removal of all real constants to an initialised COMMON block. This lengthens the code required to access them, but uses only one word of lower data instead of two. More work on reducing lower data usage is needed, however, since the provision of a lot of subroutines leads inevitably to the use of a lot of lower data (one word per argument of each CALL).
The contour plotting package that John Lewis obtained has been modified to run under SPROGS with the minimum of user inconvenience. It has been extended to allow any shape boundary region to be specified, and to allow a user to determine which contours he requires rather than drawing an even number of contours spread across the range of values.
The initial script for the film has been specified, and work is progressing on various parts of it. The decision to use as little text on the screen as possible has led to some hard thinking on the best way to present the features of SPROGS in purely visual terms. No doubt this will entail frequent re-thinks, and the scrapping of many ideas, since there is very little graphic expertise available in the design field.
Programming this film has already led to a number of modifications, and to a re-appraisal of the priorities of new features. It seems likely that local variables will be required in the near future, as will a better language than FORTRAN to specify the film! The form of this language, and that of the attendant DATA mode (input of SPROGS commands at run time), are currently under discussion.
The SPROGS system now includes the facility for storing picture files in libraries which are GEORGE files. This has enabled the picture files for the character fonts for SPROGS to be stored in some special libraries. This saves a considerable amount of space for each run of the SPROGS program since the source of the picture file definitions of the SPROGS fonts no longer needs to be included in the consolidation. The time to retrieve a font is about the same. In the old system the picture file definitions had to be executed and in the new system the picture files themselves are read from a library in the GEORGE filestore.
The library system has now been documented and this has been included in the filestore version of the SPROGS manual.
The SPROGS macro is being altered to make use of the TASK macro. This has considerably shortened the macro, it is now less than a quarter of its original length. However some problems have been encountered. In order to make sure that the SPROGS semi-compiled library is the last parameter for the consolidator the user's parameters have had to be included in brackets. The JT, MB, and MEDIA commands have to be picked up before TASK is called since they are ignored when they occur inside brackets.
A new TASK parameter *SPROGS has been included by GWR to enable the SPROGS default program description to be used if the user does not supply his own. Also certain SPROGS parameters such as NOTAPE are not recognised by TASK and cause the macro to fail. These therefore have to be checked for and are modified by adding a ? which makes TASK ignore them.
A routine has been written which will output the user name on a separate frame at the beginning of a SPROGS film. At present this is achieved using the hardware character set but the characters are too small to be read on microfilm. Alternatively, the previously defined character fonts could be used but this would take too much space.
The routine uses SD4020 instructions, and each character consists of as few lines as possible. This does mean that some of the characters are not too exact, but they are sufficient for identification purposes.
Some routines are being developed for reading plotter tapes into the 1906A, mainly for communication between one package and another or one machine and another. One problem is that GEORGE will not ONLINE a specified worktape. Therefore it seems that tapes presented to the subroutines must be owned tapes or non-standard tapes.
Ideas for the routines are presented in SPROGS Paper 35.
The main effort during the last quarter has been directed towards producing an off-lining system for small graphics jobs. In this system, any graphics job not specified as LARGE will have its plotter output go to a disc file, namely the file ACLGRAPHICS on disc 35, instead of to a magnetic tape. Each job is restricted to 502 data blocks in this file; it is hoped that this figure is sufficiently large to encourage users to employ the off-lining system for short jobs, and hence save on the number of magnetic tapes required for graphics jobs.
A job to plot out the contents of this disc file onto one magnetic tape (the file will hold the output of up to 25 jobs) has also been written, and is held on the directory :OPERATORS, so that to plot out the disc file, it is only necessary for the operator to type the command, DO PLOT.
This system will be introduced initially for the GROATS and SPROGS systems, but it may be extended to the other graphics systems on the 1906A.
Other alterations to the GROATS package include a tracing facility, to trace GROATS procedure calls: this is only included in a program if the user specifically asks for it, since this facility degrades the performance of the system.
Further, slight alterations to the GROATS macro and one GROATS procedure have enabled the system to be used with a FORTRAN master program. Documentation on this will be issued when it has been written.
The following is a chronicle of machine faults since 1 July 1973.
Outstanding Problems at 1 October
In order to demonstrate the more subtle machine faults and keep track of repairs in progress, JRG and WDS estimate they have spent 3 man-weeks of this quarter at the machine with the DEC engineers. On 16 July a letter was sent by FRAH to DEC Reading requesting a concentrated effort to correct 12 faults existing at that time. As a result, it was discovered that ECOs were not being performed as they were received. In fact, the entire VT04 backplane was replaced in August because this seemed simpler than determining the update state of the hardware.
DEC has promised to set up a schedule to perform ECOs in the future. Talks with staff from other PDPI5 sites hint that maintenance is more regular in other field service areas. It is suggested that DEC Reading be pressed to begin regular maintenance (bi-monthly) and equipment update (scheduled at our convenience) on the ACL PDP15.
We have received from DEC the tapes for an RSX system. Not much time has been spent on this yet. Our final system will in fact be RSX3 a development of this and will be compatible with DOS our current system.
The regulated power supply for the synthesizer interface has now arrived. Some final design work on the instrument panel for the interface has been done and the remaining plugs, sockets etc have been specified. RHEL are now making the instrument panel and will be installing it in the interface unit.
Programs for retrieving new lists of commands (menus) from the disc have been written but not debugged. A short film was produced demonstrating an early version of the system.
ICL were in a position to start testing the BSI package of the Mark 7 Executive in the middle of July. It was tested back to back at ICL using RBSA and by August, the Executive was ready to have the PDP15 attached. On 4th August, 2 ICL people (including an executive writer) and JRG tested Executive with the 2 computers. As fate would have it, a hardware fault appeared in the 1906A side, but was not recognized as such (since Executive was suspected first) until Ted Evison ran an executive-independent version of RBSA later in the day.
By the end of the week all the Mark 6 link programs were running and transferring information.
The following week (14th August), the BSI caused 2 machine breaks because of confusion over Executives. For a period of time it was difficult to find out from anyone what Executive was being used. The trouble appeared to be lack of communication between engineers going on and coming off holiday.
The present state is that an Executive mend is being awaited from ICL to prevent Executive causing an auto-PM on a certain hardware condition.
This is in the same state as the last QPR.
Next year, the COLAB management game is to be incorporated into two courses. However, neither of these courses will be held at Cosener's House, and so a new organisational procedure will have to be adopted. Since the course centres only possess one external telephone line, it appears that it will not be possible to allow the players direct access.
This means that a single user version, where Control types in the data for the players, will have to be provided as an extra option. A new version of COLAB using a conceptual multiplexor will be produced and tested during the next quarter and this version will contain the single user option.
A paper by RET and DCT describing the various COLAB implementations has been submitted to Software Practice and Experience.
The complete set of Hash Table Films has now been produced and final prints of all films should be available within the next two weeks. The set of films are:
All five films will be made available to universities as teaching tools. One university has already shown interest in purchasing a set.
PIGS - PDPI5 Interactive Graphics (for DECUS 73 Proceedings), W D Shaw and J R Gallop
Some Aspects of the Production of Computer Generated Movies for Chemistry Teaching (for International Conference on Computers in Chemical Research and Education, Ljubljana, Zagreb, 12-17 July 1973), F R A Hopgood (with P Groves, University of Aston)
