This is the last issue of FORUM for 1981 and I would like to send Christmas greetings from the Computing Division to all our readers and users.
This year has been one of considerable change and, always, that tends to have adverse effects on the user service. However, there are clear signs that the front end service on the 3032 is improving. The move to CMS during the year has been quite dramatic and we are pleased that we have been able to sustain the good response on CMS despite the ever increasing user population. This has had a second advantage in that ELECTRIC performance has also been improved. The programme of installing multi user systems in university departments for interactive working is now just about complete and I anticipate that the major changes in this area are likely to be enhancements of existing systems. Bringing the funding of both the batch and interactive services together under the Central Computing Committee has allowed us to be rather more innovative in our enhancement programme. We are hoping to upgrade some of the GEC systems to the new 32-bit GEC 4090 and this will allow us to replace some of the 2050 workstations by newer GEC 4000 systems.
I anticipate that 1982 will also be a year of change. We are optimistic that some part of the plan for enhancing the front-end facilities and replacing the 3607195s will be achieved. On the network side, a number of new packet switching exchanges are being installed at major centres (such as London, Cambridge and Edinburgh) which will improve the overall performance of the network and provide greater rationalisation of leased lines. It is likely that some long term plan for a combined SERC/Computer Board network will be established.
A third disk has been added to the facility for short lived data sets. The proposal mentioned at the June representatives meeting (see FORUM 14) is now partially implemented. When creating a short lived data set, users may specify
DISP=(NEW,CATLG),VOL=REF=FREE,DSN=DEC.ID.EXAMPLE
and when subsequently using the data set
DISP=SHR,DSN=DEC.ID.EXAMPLE
It is important when using existing data sets to omit the VOL=REF=FREE clause. The reason is the definition of REF=FREE is time dependent. It will be changed according to the availability of space on the three disks USDSK1, RHEL08 and the newly created disk FREE03.
Please discontinue the use of data set names of the form USER.NOV for short lived data sets. They will cease to be legal after 1 August 1982.
At the time of writing, CMS accesses, ELECTRIC OS copies and FTP access still require explicit volume references. Some work has to be done to overcome this.
The new disk, FREE03, is a model 3350 equivalent. The older disks, USDSK1 and RHEL08 usually referred to as FREEDISK and ATLAS, are due to be copied to 3350s in the near future, hopefully by the time this article is printed.
It is now possible to run this compiler in a user's own CMS Machine. (FORUM 16 explained how to use it in CMSBATCH). The command FQRTVS may be used in the same way as FORTGI i.e.
FORTVS ABC
will compile the source found in ABC FORTRAN to create an object module in ABC TEXT and a listing in ABC LISTING with a summary output coming to the user's terminal.
There is no further information available on the availability of optimization.
The compiler may only be used under CMS but the object modules may be used in the MVT system provided that they are linked with the FORTRAN library routines contained in SYS1.VFORTLIB.
The latest version of the NAG library has been implemented and is to be installed in December. The load module and object module libraries for MVT will be SYS1.NAGLIB and SYS2.NAGLIB respectively. The NAGLIB TXTLIB available in CMS will be replaced at the same time. The Mark 7 versions will continue to be available for a short while as SYS1.NAGMK7 and SYS2.NAGHK7 in MVT and as ONAGLIB TXTLIB R in CMS. These will be removed early in the new year.
Recently there have been some changes to the mechanism for submitting jobs to HASP from a job running in the MVT batch. There are two side effects. A minor effect is that job submission does not occur under the new mechanism until all output has been 'printed'. If the job is routed to a workstation controlled by VNET, this in practice means as soon as the job ends because any queueing for printers etc affects VNET and not HASP. If the job is routed to a workstation controlled by HASP then any delay in printing the job will also delay the job submission. Clearly the conversion of workstations from HASP to VNET will reduce the impact of this.
The second effect is that, for the present, using the new mechanism it is not possible to use the internal reader from any job which produces FR80 output. The problem will be overcome when a new despool mechanism is introduced. Work is in progress on this hut it is not possible to give a date when it will be introduced.
Any user who is seriously inconvenienced by the recent change should discuss the problem with the program advisory service. It is possible to force the old mechanism to be used, but it is not considered desirable to perpetuate it.
Many users may imagine that they never use the facility anyway. In fact the library cleanup mechanism is triggered by the submission of a job through the internal reader each time a user library expands into a new extent.
8 PACX ports have been transferred from ELECTRIC to CMS. This is in accordance with the plan to run down the ELECTRIC service as CHS use increases.
At last the new version of the VM Reference Manual has gone to the printers. The main changes are:
Section F - This describes XEDIT instead of LDIT. No details of EDIT are included in the new manual.
Section E - This now describes EXEC2 as well as EXEC.
Section G - New section on VNET.
Section R - Many new commands.
Section S - New section, containing 'commands for 3270 full screens and magnetic tapes, which are not generally available. This section will only be included on request.
The material in Sections R and S is also covered in the HELP files. Copies of the individual chapters (A - G) and the index are available on USDOC 193 disk (read password RUSDOC), with filetype CHAPTER. These can be printed by you (remember to request carriage control using the CC option).
It is our intention to send copies of the new manual to all registered users as soon as they become available. This may take a little time.
Comments on the content and structure of the manual are always welcome and will help us get it right next time.
PATCHY was upgraded on 23 November 1981 to version 1.04/8. A few changes to the procedures are in hand. The ALERT/NEWS files should be consulted for details.
Version 3.32/1 of HYDRA has been installed and is available now. The documentation lists for PATCHY and HYDRA have been combined. Documentation for the latest version of HYDHA was sent to all those on the combined list on 16 November '981. Please contact June Scholes, RAL ext 272 if you have not received a copy and wish to receive the current and/or future documentation on PATCHY and HYDRA.
Please note that the older versions of HYDRA will be purged from the central computers early in 1982. Contact Barrie Whittaker, RAL ext 6674 if this will cause difficulties.
Up-to-date information on PATCHY and HYDRA is available on the central computers through the ELECTRIC files DOC=ALERT.PATCHY and ALERT.HYDRA and in CMS through NEWS PATCHY and NEWS HYDRA.
We are often asked by potential grant holders how to convert time on their local machine to 360/195 hours. As an aid to conversion, the following figures are provided as a guide to the amount of processor time required for batch jobs. The ratios are derived from Central Computer and Telecommunications Agency (CCTA) synthetic benchmarks for single precision FORTRAN, manufacturers HIP (millions of instructions per second) figures and other relativities, each representing some rather artificial or arbitrary measurement conditions; hence, wide variations will be found from application to application. The figures should not be used for general comparisons of the different systems.
On examining the figures, you may find from your own performance information somewhat different ratios, especially as the numbers given are relative to the 360/195 which has a fairly wide performance range. We would be interested to learn of any such information, so that we might update our own tables.
| SYSTEM | × 360/195 |
|---|---|
| IBM 360/195 | 1.00 |
| IBM 4331-1 | 0.03 |
| IBM 4341-1 | 0.20 |
| IBM 3033U | 1.10 |
| AMDAHL 470/V8 | 1.50 |
| BURROUGHS B 7750 | 0.16 |
| CDC CYBER 170-20 | 0.18 |
| CDC 7600 | 2.00 |
| CDC CYBER 205 (2 PIPE) (SCALAR) | 2.40 |
| CDC CYBER 205 (2 PIPE) (VECTOR LENGTH 1000) | 42.00 |
| CRAY 1 (SCALAR) | 3.20 |
| CRAY 1 (VECTOR LENGTH 64+) | 20.00 |
| CTL MODULAR 1 | 0.01 |
| DATA GENERAL ECLIPSE S/200 | 0.10 |
| DEC KL10 | 0.24 |
| DEC PDP 11/45 (FPP) | 0.04 |
| DEC VAX 11/780 (FPA) | 0.24 |
| GEC 4080 | 0.05 |
| HARRIS S 500 (SAU) | 0.16 |
| HP 3000-III | 0.04 |
| HONEYWELL 6/43 (SIP) | 0.05 |
| HONEYWELL 6080 | 0.16 |
| ICL 1904S | 0.04 |
| ICL 2960 (VME/B) | 0.08 |
| ICL 2980 (FMDU) | 0.72 |
| ICL SYSTEM 4/72 | 0.06 |
| MODCOMP CLASSIC | 0.19 |
| NORD 50 | 0.11 |
| PERKIN ELMER 3220 | 0.11 |
| PRIME 750 | 0.20 |
| SEL 32/77 (HW FL PT) | 0.12 |
| SIEMENS 7748 | 0.06 |
| UNIVAC 1108 | 0.16 |
The argument passing mechanism used by all the PRIME, GEC and DEC10 FORTRAN compilers is very simple. Arguments are all passed by reference, ie the address of each argument is passed to the subroutine, so no local storage is allocated for dimensioned or non-dimensioned variables. The subroutine accesses the original storage locations.
As with IBM FORTRAN, this allows the passing of the dimensions of an array as arguments which can be calculated at run time. The possibility of a particular variable being overwritten by a subroutine can be eliminated by passing it as an expression, for example by enclosing the variable in brackets. This will cause the address of the temporary variable used to hold the result of the expression to be passed to the subroutine, rather than the address of the variable itself. Of course, this should only be done with non-dimensioned variables!
A one-day speech recognition workshop is to be held at the Rutherford Appleton Laboratory early next year. Included in the programme is a demonstration of the LOGICA speech recogniser, LOGOS. Would anyone who wishes to attend please write to: Dr I D Benest, Rutherford Appleton Laboratory, Atlas Centre, Chilton, DIDCOT, Oxon, OX11 0QX for an application form.
List of articles in FORUM 15, 16 and 17
15.1 The CMS editor, XEDIT 15.2 Questions raised at CCR Meeting, 13/7/81 15.3 Index 15.4 Correction to RAL computer newsletter 9/81 15.5 Computer statistics 15.6 Network HELP 15.7 CMS courses 16.1 Introduction 16.2 Changes at RAL 16.3 Central Computer Procurement 16.4 IBM VS FORTRAN Program 16.5 Extract from minutes of PRIME User Meeting 16.6 Questions raised at CCR meeting, 13/7/81 16.7 Courses 16.8 Diary 16.9 Any offers - information needed about tapes 16.10 Computer Statistics 16.11 Index 17.1 Introduction to VNET 17.2 Some notes on UNIX 17.3 EDIT to XEDIT migration 17.4 File protection 17.5 Computer Statistics 17.6 Diary 17.7 Correction to Item 5 in FORUM 16
20 January 1982 - IBM Representatives in RAL Lecture Theatre
The programme for this meeting includes items on mainframe procurement, initial thoughts on MVS, some aspects of operations, progress reports on charging and control, VNET and networking as well as a general session.
March 1982- Prime user Group at UMIST
Thursday 24/12/81 - Christmas Shutdown (IBM System) at 1600 hrs
Tuesday 29/12/81 - Computing Service (IBM System) reopens at 0800 hrs
