ILL - Central computer evolution

Until 1973, the ILL access to a mainframe was through a remote job-entry terminal (RJE "terminal lourd") connected to an IBM-360 computer at the computer centre (IMAG) of the Grenoble university. This offered a card reader, a card puncher, a printer for batch jobs (i.e. scheduled jobs, not interactive jobs).

A few Teletype ASR33 terminals also offered some kind of interactivity through CP/CMS (Control Program/Cambridge Monitor System) but not much appraised since their respective time-sharing and batch FORTAN compilers were incompatible.

(More about it in page "Main system replacement")

Users were encouraged to take away data even when the instrumentation and data treatment packages were specific to the ILL.

Until Mössbauer (1972-1977), there was no Central Computer only a remote connection (RJE) to the University mainframe for card input and printer output, though a few terminals were available too. Unfortunately the compilers used by the interactive system (CP/CMS) were incompatible with the OS/360 compilers.  In 1973, the RJE was replaced by a modern DEC PDP KI-1070 (also known as DECsystem 10 or DEC10 or simply PDP10) a 36 bit computer. The French director Jacrot risked his career to implement  the ILL choice since the CEA was fiercely against it.  The DEC10 was an excellent computer, so easy to use that physicists could start it up themselves during week-ends without the shift operators.  Once installed there was reluctance from the Management to enhance the computing power though crystallographers at the Grenoble CNRS group appreciated the time-sharing facilities it offered.

To aid the scientific program a small group, GRATIN was created (Filhol, Hueskin, Ghosh) with knowledge of activities in Diffraction, Nuclear Physics, and condensed matter physics respectively.

A more heavily political choice was that to replace the DEC PDP KI-1070, by 1980/81. Already over seven years old it was heavily used by all groups. This project is described in more detail in this page. The new computer, a DEC KL-1091S was finally installed in a new computer building in 1981, and was essentially a faster version of the original KI-1070, with better networking front end processors, and new tapes (6250 bpi), Printers (1200 lpm), etc. Some of the KI10 peripherals were retained and reused. Optimem, 12 inch re-writeable optical disks were used for data archiving up to 3 Gb of data, essentially tape image files, reducing pressure on the 9-track tape drives.

The introduction of a VAX8600 in 1986 to take over from the DEC10, then offered complete compatibility with 32-bit instrument computers, again using the ethernet local area network with DECnet throughout. The VAX was popular with scientists, and to aid use by the ESRF the 8600 was upgraded with a financial contribution from the ESRF, to a VAX8650. The actual configuration was a cluster including an HSC50 storage controller, and a VAX11/751 (recovered from the upgraded PDP11/55 NICOLE replacement concentrator). It was hence easy to expand in 1988 with a VAX8700 cpu joining the cluster.

In 1992 the ILL Steering Committee decided that the Institut should break the monopoly of DEC systems and replace all with Unix systems.  During the long shutdown the nuclear physics groups and the inelastic instrument groups were given HP-PA 9000-730 RISC systems and other groups were given Silicon Graphics RISC systems.  A central Silicon Graphics server would replace the DEC cluster over time. For the UNIX systems (and compatible too with PC-Windows and VMS) it had been decided that the data format would be in ASCII, based on the ILL-TAPDAT extensible format.  Compressing these data before storage allowed the OpenVMS, Unix and Microsoft systems to regenerate files matching each systems characteristics.  Space was saved, and network traffic reduced.  This also resolved problems of the different machine binary configurations.

When the reactor was restarted in 1995 there had been little renewal manpower or money for instruments. Most instruments were still running VAX-VMS. Data were transferred to the central cluster and converted to ascii before being written to the SGI unix server.