There have been quite a few changes to the Group's membership during the last quarter. Alex Rowson left at the end of August to move to Colchester. In addition, the two students, David Ralphs and Peter Davies, returned to college. The need to proceed as quickly as possible on the UK4 data processing has meant that Anton Walter has moved to the Applications Software Group in order to devote all his time to this project.
Until this quarter, the Group has shared with the Application Software Group, the services of Dianne Byfield as Group Secretary. We now have our own secretary; Judy Chamberlain. The Group has also recruited Frankie Sims and John Thewlis during the quarter. In addition, a student, Peter Hayes, from Portsmouth Polytechnic, will be with the Group for a year. The only other recruitment in the pipeline is Graham Robinson who arrives on 3 October. The staff will then be:-
We were fortunate in having Prof Schwartz from the MIT Education Research Centre visit the Laboratory for a week in August.
During the period two releases of GEORGE have been received from ICL, 6.5 and 6.6. These releases and other MENDS received have cleared up a number of important bugs which had depressed performance in the past. There are still a few bugs which are hindering the machine and progress. In particular, a bug in the High Level Scheduler's saving of jobs is causing an incident if the WHATJOB is active at the same time. There appears to be a bug in trusted programs which is hindering development of COLAB. In addition, a few user activities are known to break the machine. However, the state of GEORGE is satisfactory in that the break rate is only about one or two a day if multiple breaks caused by a user continually running particular programs are ignored.
It is understood from ICL that no further GEORGE 4 Mark 6 releases will be made before Mark 7 is released in December. It is, however, hoped that the few irritating bugs which are holding up progress will be cured before then.
The hardware has been fairly satisfactory. From time to time a bad day is encountered which results in a number of GEORGE breaks. The 7903 has been a continuing cause for concern since it appears to give trouble at least once a week. The fixed disc was out of action for a number of weeks but is now in use again. The fault which causes the device to appear unavailable to GEORGE is still present.
The GERONIMO visual displays have now been delivered and are being installed by ICL. The executive for the system has been ordered but the GERONIMO code has not arrived. GERONIMO documentation is now to hand.
Four Racal Milgo modems were delivered on 18 August and the three rack-mounted ones have been put in the racking. This required some metal-work due to incompatible dimensions. The Belfast line was converted to Milgo and has operated satisfactorily since then. Curiously enough, all the new modems have blown fuses, but it is hoped this is not a feature of these devices. The CASE modem has been returned and will shortly be tested once the cables have arrived from Belfast.
The 1200-based link to Imperial College is not yet a success. Initial attempts to use it have failed. SIGMA 2 is now being programmed to attempt to test the interface to the 1900. Initial tests have proved negative although success was achieved when driving 110 baud links using SIGMA. The basic problem is that the 7903 does not cater for 1200-based teletypes and code changes have been made to attempt to alter this. These changes do not seem successful yet. Unfortunately, due to ICL removing all reasonable means of modifying the 7903 code, this has to be done by hand switch manipulation which is tedious and liable to result in errors. The project is unlikely to be a success until ICL provide an easy means of altering the code.
Negotiations have continued with GEC with a view to supplying five RJE terminals based on the 2050 computer. These will be similar to the RJE terminals being acquired by RHEL. ACL will provide 7020 emulation. It is hoped the machines will be delivered by 1 April. Some ideas for using one as a front end to the 1906A are being considered including the possible provision of a standard 1900 interface on the 2050 to allow it to imitate the 7903 for high speed communications.
A hardware monitor is being constructed to measure the fraction of 1906A time spent in user programs and in EXECUTIVE. The hardware is substantially complete and is held up awaiting a cabinet to house it. The device has been designed here, based on an idea from Leeds University.
Leeds University have used their monitor on the ACL machine to measure the effect of more core than their 128K on MOP performance. This is reported. in Technical Notice 7.
During the quarter backing store jams have become more frequent. Initially, the parameters were set to give a jam once a week. This caused the jams to be severe. On the advice of, a paper by Ian Battersby of ICL, it was decided to attempt to cause jams more frequently and throw off less material. It had been observed that the unjammer tended to throw off large files. The theory is that it is more economic to throw off large files but ACL observed that small useless files could be removed with advantage and thus, hopefully, fewer large files, or even just fewer files, would be retrieved subsequently. The unjammer size-dependence was therefore removed and unjamming depended only on when the file was last accessed. To measure the effect of the unjammer a performance package was mended in and also the number of retrieves every five minutes was put in the performance file. This work was partially successful in that the retrieval rate did not rise and a large number of small files were removed. Unfortunately, since each residence is unjammed independently, it is difficult for FORMULA to adjust itself to cause a jam every two hours which was considered reasonable.
The second investigation was to unjam all residences over, say, 85% full when anyone jams at 90% and to attempt to unjam to something like 75% to 80%. This work is still progressing and will be reported more fully later.
UK4 has shown that the Low Level Scheduler is not satisfactory. The algorithm takes into account:-
The algorithm does not identify peripheral-bound jobs. UK4 is just such a job and if it shares with a mill job, the mill job depresses the 1½ hours of a UK4 job to perhaps 7 hours.
This problem has temporarily been cured by making slot time small. This is beneficial since it causes jobs to be switched more frequently by the fair shares for all theory of the LLS. Thus, since UK4 takes little mill time, it will normally be halted in its slot and will not have to wait long when free to run again.
A better algorithm which attempts to schedule more on the periperal-boundness of a job is more suitable for ACL.
Currently, the LLS code is being examined. Unfortunately, this is not an easy area to investigate. Moreover, it is not easy to detect whether a job has been held up for a peripheral transfer.
A series of GEORGE amendments have been produced which lead to the automatic abandonment of timed out MOP jobs after a few minutes. Currently, large MENUDS being worked on are
A copy of GEORGE on microfilm has been produced since a lineprinter listing is large, bulky and gets tatty and torn quickly. Various other ideas of microfilming other parts of GEORGE are being considered and also whether to sort the chapters into alphabetical order on the microfilm.
A framework HLS, which simulated the limited job scheduling methods that took place previously, was put into the standard system in the middle of August. From this point developments could normally be tested at any time instead of during special system sessions. The EXPRESS command can now be used, and certain operator information is given (in particular, the entry of long jobs into the system).
More sophisticated job scheduling is implemented and tested but its use is held up by a GEORGE bug connected with the saving of jobs. It is hoped that when we receive the mend for this, the saving scheduler can be switched in and joblimit made very much higher. The effect of HLS should then really be noticed especially by batch card users (with short jobs) who currently get a very poor service during times of heavy load.
Some limited job statistics (eg number of jobs processed) will shortly be output to the performance file by HLS.
Evidence has been produced to illustrate and correct various causes of inefficiency in the running of the 1906A. These are as follows:-
The performance package has been modified to print out the installation parameters and various kinds of messages.
Resource allocation has requested a program to cross-reference the identifiers in a FORTRAN program. The project is being based on a FORTRAN tidying program written in PLASYD by A J H Walter.
The program will provide identifier lists and cross-referencing similar to those produced by the 360 compilers.
The new COPYIN and COPYOUT programs are being examined and a paper on the exact difference between the new and old versions will be produced.
This work may be expanded to consider the archiving of the information on an old dump process tape.
Although JDT only joined the Group in September, he has previously been involved in work for Resource Management Section. This work was involved with reading 1108 magnetic tape and the development of a MACRO for simulating the LISTDIRECTORY command but including a form of recursion to list the directories of any pseudo-users found.
A considerable amount of work has been put into a 1906A implementation of COLAB, much of it purely experimental (due to the lack of documentation on some of the ideas mentioned below).
On ATLAS, the COLAB game is run largely on SIGMA 2, with one controller and up to four players: these communicate via the message facility in the SIGMA system, and the game progresses via a sequence of independent FORTRAN jobs submitted to ATLAS.
On the 1906A, it is hoped to run the, system interactively: the controller and players will each have a program on-line to the corresponding console, and these programs will communicate via basic peripheral communication files. In addition, the control program will be a trusted program and this will run the FORTRAN programs (as run on ATLAS) as programs under control. Under this system, requests for contract bids or other data may be queued by the FORTRAN program, which can then check the validity of the data, and ask for a repeat if it is not correct. It is hoped that the resultant systems will give a good turn-round time and will be generally faster than the ATLAS/SIGMA 2 system.
While all of the above ideas have been tested and proved feasible, the latest version of the control program running under the latest versions of GEORGE (GEORGE 4 Mk 6.6 versions E and F) causes a system break: the reason for this is not yet known.
Most of the quarter has been spent in discussing a number of outstanding problems, trying solutions and keeping the manual up to date (a nontrivial problem!). Hardware character plotting routines for both lineprinter graphical and SD4020 have been written, together with an initialising and clean ending system. Hardware character numbers have changed more than once to accommodate the changes to the storage of text in the system. The old scheme of a string table proved impractical, due main1y to FORTRAN's inability to handle text nicely, so text is now packed into a real number form.
The format of files and the inclusion of data in files by the use of NULL routines have also changed. However, it is hoped that the storage of fi1es, at least, will be implemented in the next few weeks. A system was written to deal with NULLS, but again this has had to be thrown away. Their implementation will be delayed until simple files can be shown to work.
An outline proposal for a user trace facility has been put forward in SPROGS Paper No 19. Exactly what ought to be printed will be shown as the system is used. Currently, the lineprinter output mode provides some form of debug output. DR attempted to produce a film sequence in SPROGS, based on a drawing by FRAH. This managed to use 20 regions, and take 6 seconds per frame of 1906A CPU time.
When the basic line drawing plotting routines were written, tests were carried out to discover how fast the system was running. To assist, the ICL FORTRAN tracing package was implemented and a test picture run through it. This uncovered one source of inefficiency which was removed. The final version showed that the system was running at 350 lines per second (lines here being SD4020 short vectors). This is, however, about half the speed that GROATS achieved. It is probable that the extra time is taken by SPROGS running in EXTENDED mode, and by the relatively poor code produced by FORTRAN (as opposed to ALGOL).
The SD4020 routines written for GROATS and used for CAMP and CAMPER have now been interfaced to the main SPROGS program. These output routines and some of the interfacing routines are written in PLAN but since they are for a specific output device they do not contradict the concept of SPROGS being machine-independent. Vectors and hardware characters can now be plotted on both hardcopy and microfilm. Some more routines have been written to obtain from GEORGE the username, jobname, time and date of the run. These. are then output on the first few frames of the hardcopy and microfilm along with a cut frame, the laboratory slide and a title. The magnetic tape numbers used and other information for the SD4020 operators are output to a lineprinter file which is listed at the end of the run. The initialising sequence is performed the first time that either hardcopy or microfilm output is selected by the main SPROGS program. It currently uses hardware characters but will be updated to use software characters when they become available.
Work has started on designing a library system for SPROGS picture files and film files. A standard disc file will have a directory of the film files it contains as well as the actual files themselves. There will be two types of library files, a standard set which will contain the software fonts held as picture files and some standard film files, and the users' library files where a user can keep film files which he wishes to use in future runs.
Work in this quarter was divided largely between the stylus input package, STYLII, and definition of a command handling system for PIGS (PDP INTERACTIVE GRAPHICS).
A grid mode for inputting points was added to STYLII. The lightpen tracking routine was re-written (as LTPEN) to make it look more like the tablet. Documentation is available in SPROGS Paper No 20.
At a meeting in August, a suggested PIGS command system was detailed. The package, as defined, would include a command structure definition language and run-time command handling and creating facilities. The package is flexible enough to be used in other interactive PDP systems. A paper will be produced as soon as the actual commands to be included in PIGS are determined.
A paper Automating Animation was delivered at ONLINE 72, describing the SPROGS system. It was necessary to prepare a set of 32 slides to accompany this paper, and various methods were tried:
The slides made by method (b) were eventually used in the talk, RHEL producing white-on-blue photographs.
The paper will eventually be published in the proceedings of the conference.
The problem of the ICL 1906A BSI not interrupting the processor was due to the program in the 1906A being started too late. The Executive of the 1906A threw away the flag from the BSI as no program owned the device. When the 1906A program was started before the PDP15 program, the link worked.
Characters were packed and unpacked on the two computers in opposite senses. A message in the PDP15 that read :-
ABC DEF GHI JKL
was read by the 1906A as:-
CBAF EDIH GLKJ
This was solved by moving two plugs.
The link now works at a data rate measured at 60K baud. Further timing trials will be carried out to locate why the link is 20 times slower than expected.
Text 360 has been implemented on the IBM 370/195 at RHEL. A User Notice will be produced using this package to describe its operation. Also a more readable and publishable form of the SPROGS' manual could be produced by it.
A short film in colour showing one method of constructing a cardioid has been made using GROATS. The program was written, depending on three variables (representing three colours, A, B and C). Four films were produced, three containing information for only one colour, and one being the total sequence in black and white. This latter film was used as a mask to produce a coloured background, while the three separate films were processed in different colours and superimposed on this background. The final film is one of a series which will be made into film loops.
A paper has been written and submitted for presentation to the Users of Automatic Information Display Equipment 1972 Conference. Some slides have been drawn and are being made for this conference.
Whilst in North America, visits will be made to the following establishments to discuss their computer graphics work.
CAMP and CAMPER are both now working on the 1906A and a note has been issued describing the differences from the ATLAS version. CAMPER is also now working on the PDP15 (CAMP having been completed last quarter) and a note describing its operation has been issued.
A new error trap routine has been incorporated into the 1906A versions such that, in the event of an error occurring or run-time being/exceeded, the output magnetic tape will be correctly closed and some diagnostic information will be output. A new parameter has been added to the macros for running the programs allowing the run-time to be varied. A new CAMP/CAMPER command has been added to request retention of the SD4020 magnetic tape for a number of days before it is returned to the scratch pool. The default setting is now such that the tape is returned to scratch immediately' after processing.
A new comprehensive manual for CAMP and CAMPER has been written. It will replace the existing manual CAMP and CAMPER on ATLAS and several notes issued to describe use on the PDP15 and 1906A. The new manual, which is now in the process of being typed, will describe how to use CAMP and CAMPER on all three machines.
FOCAL is an interactive language for evaluating expressions and writing simple programs. External functions can be written but these can only extract arguments by value.
A small argument processor (currently being written) will enable a user to refer from FOCAL to an external FORTRAN callable routine.
Argument references can be used as well as simple values; references to COMMON will be implemented also. The mechanism for transferring an array name is still being considered. This will be useful for testing FORTRAN callable subroutines and for producing programs in a short time this will essentially link together subroutines already produced.
The 1906A GROATS system is complete and working, and, apart from mends for two bugs, has been stable for most of the last quarter.
Work on the manual is still in progress but this should be completed before the end of the next quarter: the first seven chapters of the manual are available for any user who would like to use GROATS on the 1906A.
The GEORGE filestore directory containing GROATS, namely :GSIN00, is fairly long, and as part of a drive to improve the efficiency of GEORGE, this directory will be split into a number of pseudo-user directories; users of the GROATS system will be informed as and when this affects them.
A formal presentation of the Texas Instruments' ASC computer was given to the Laboratory in August. As a result of this, an evaluation of the machine was prepared as an addendum to the earlier Machine Evaluation paper.
A system has been produced for the definition of slides and film titles on the PDPI5. The user makes up the relevant information on the display and, when satisfied, generates a SD4020 tape containing the desired information. The information displayed is basically a set of characters but more complex pictures can be displayed by redefining certain characters as picture objects. A part of the system is a package whereby the user can input new character forms using the lightpen. This is useful for defining new character fonts in GROATS or SPROGS. A complete definition is given in SPROGS Paper No 17. The system is still unable to generate slides consisting of two 35mm frames abutted, due to the continued inability of the SD4020 to perform with the required accuracy.
SOFTWARE 72 24-26 July - University of Kent, A J H Walter and A G Rowson
ACTP Conference 21 - 25 Aug - University of Surrey, P E Bryant and C J Pavelin
ONLINE 72 4 -7 Sept - Brunel University, R.E Thomas J R Gallop F R A Hopgood
DECUS Europe 19 - 22 Sept Strasbourg, G A England
Computer Science Colloquium, Edinburgh 19 - 22 Sept, D C Toll
ICLCUA Autumn Convention - Reading, 20 - 22 Sept, P Kent
