The last two months have been very busy for Central Computing Division staff at RAL. As well as preparation of the software for automatic archiving of CMS minidisks and the tailoring and testing of CMS Release 4 there has been a lot of activity connected with the Cray.
The X-MP/48 was delivered on December 15 and was quickly installed by Cray. All the vital site preparation - cooling, power etc - having been prepared by RALs own expert engineering staff.
Acceptance tests were followed by installation and tailoring of software. Invited users were able to use the machine by 14 January and a full service was available by 2 February.
The X-MP/48 service. now run under the title Joint Research Councils' Supercomputer Unit, has its own newsletter but details of available software are included in this issue of FORUM for general interest.
Installation of the Cray has necessitated the implementation of JTMP on the front-end IBM 3081. Work is in hand to enable it to communicate with the CMS SLAC batch system.
Another development of interest to network users is the delivery of the first 'fawn boxes'. These implement the SSMP protocol across JANET allowing a standard terminal to work in full-screen mode to a host which is also running SSMP. Tests of SSMP connection to CMS are under way.
The IBM MVS service still has spare capacity at night and weekends. If you are planning bulk production work. now is the time to submit it.
The Engineering Computing Facility will continue to contribute to FORUM those articles which would be of interest to our readers. despite producing their own Engineering Computing News. In this issue they announce a new Initiative in the Engineering Applications of Transputers.
Jim Diserens' article shows that Electromagnetic Software can span both PR1MES and MVS using the same package.
The Masstor M860 Storage System is a large capacity data storage device, the largest model of which is capable of storing in excess of 200 Gbytes of data in such a way that any item of data can be retrieved in about 20 seconds. It achieves this by writing the data on a length of magnetic tape 66 ft long and 2.7 ins wide using a helical-scan rotary head similar to those in domestic video recorders. The read/write head assembly rotates at 116 revs/s which results in a linear head transit speed of 1300 in/sec over the tape. The combination of the rotating head and the linear motion of the tape writes 27080 sloping tracks, each 6 ins long and each holding about 6.5 Kbytes of data on one strip of tape, giving a total capacity per tape of 175 Mbytes.
When the tape is not in use it is automatically rolled up and stored in a cylindrical cartridge about 4 ins long and one and a half inches in diameter. About 300 such cartridges are stored in a honeycomb structure to form a storage module with a capacity of 55 Gbytes. The entire action of selecting a cartridge, extracting the tape, threading it into one of the read/write stations, positioning it, reading or writing the data and putting the tape away again is under automatic control.
Two M860 storage modules, each with two read/write stations, are installed at the Atlas Centre, giving a total capacity of 110 Gbytes.
Masstor Systems Ltd provide software (called SMT) which runs under MVS to make the M860 appear to MVS as a group of IBM 3420 tape drives. This is being used at the Atlas Centre in conjunction with a commercial package (ASM2) to provide a transparent hierarchical storage system for MVS transient and permanent disk files. Disk data sets which have remained unused for more than a certain time (presently 10 days for transient data sets and 15 days for permanent data sets) are automatically archived into the M860. Such archived data sets are automatically retrieved and restored to disk as soon as a user job references them, so that the user need not, and in general does not, know his data sets have been archived.
There are clear advantages in extending such a system to handle data held on VM/CMS minidisks, but the lack of Masstor software to support the M860 from VM has prevented the use of the M860 in this way until recently. The necessary software has been under development at the Atlas Centre for some time, and it comprises two main components. One, called STAGER, the minidisk manager, has been a running service since December 1985. This manages the backup (and occasional restores) of CMS minidisks held on a pool of about 40 real disks to real 3420 tape drives. It is also capable of automatically transferring complete minidisks to a backing store after a period of inactivity, and restoring them transparently as soon as a LINK to the minidisk is requested, but this function has not been activated due to the lack of the VM M860 support.
The second component is a VM service machine for the M860. Because MVS requires access to the M860 as well as VM, the new VM server must be capable of supporting concurrent access from several different operating systems. It does this by intercepting all requests to mount cartridges, checking that the requested cartridge is 'owned' by the requesting system, and (if necessary) waiting until the requested read/write station becomes free. Thereafter data transfers proceed transparently. This M860 service machine has been developed by CCD staff at the Atlas Centre and was installed on 4 February. It is currently being used by SMT for all MVS accesses, and early tests of its ability to control the sharing of the M860 between several operating systems, including VM/CMS, are encouraging.
One exciting feature of the new M860 server is its ability to permit access to the M860 direct from the Cray X-MP. The final checks have yet to be made, but investigations have so far demonstrated that the Cray can successfully write to the M860 and full incremental backups of the Cray filestore have been taken in this way. Code to trap the mount requests in the Cray station and pass them to the M860 server is already in place, and COS dumps to the M860 are now being taken on a regular basis. A full hierarchical storage system with transparent data migration now seems not too far away.
On the VM/CMS front it is planned that regular minidisk dumps to the M860 will commence about the end of February. During March, data will be written to the M860, but not deleted from disk. During April full minidisk staging will be implemented for CCD staff only, to ensure all the bugs are out of the system before commencing full staging on all CMS minidisks in May. The precise date when users' data is involved will be announced nearer the time. It is anticipated that about one third of the disk space now occupied by minidisks will be released with an inactivity threshold before staging of 14 days, freeing about 10-15 drives. The only effect which should be observable to users is a delay of 20-30 secs in LINKing to a minidisk if it has not been accessed in the last 14 days. This should have a minimal impact on most users.
As a result of these developments the M860 is now seen as the main repository for long-term data irrespective of the system originating that data. This simplifies future developments, eases maintenance and reduces operational effort as well as providing a much improved service to our users.
This is a progress report on applications software installed on and planned for the Cray X-MP/48 at the Joint Research Councils Supercomputer Unit (JRCSU) in the Atlas Centre.
If there is any other software you think should be installed on the X-MP/48 then please let us know.
Already installed are: