Conversion to Unix

Back to main page       Top

The proliferation of DEC computer systems had always been highly questioned, primarily by the French. To break this monopoly a change to Unix was imposed in 1992 during the shutdown of the reactor. Very limited financial resources were available for practical implementation. An extensive study within the Departement Informatique, with input from external advisors lead to a document, The Unix Plan, to cost and implement the transformation.

Two computer systems, the Hewlett-Packard HP 700 and the Silicon Graphics SGI computers were chosen. The latter, who finally owned the MIPS processor chip manufacturer, had targetted the scientific community, taking a significant part of the DEC market for scientific as the VMS systems showed shortcomings in graphics and overall computing power. At this stage the crystallography community was developing rapidly using the advanced SGI graphics (OpenGL, etc.). The HP computers were powerful, but the software support was not driven strongly in the scientific domain.

Up to this time LAN networking used DECnet, and the DEC products DFS for file sharing and DQS for sharing print services. WAN was restricted to X-25 networks, which could also carry DECnet.
With Unix there was a shift to adding IP services to the VMS systems, especially the central VMS servers. The initial use of Ultrix on a DECstation with a MIPS processor was not extended to other systems.

By 1994 each scientific group had at least one workstation available. The Three-axis group, Time-Of-Flight and Nuclear Physics were given HP-730 systems. The LSS group and Crystallographers were given SGI-Indigo workstations. An SGI- Challenge-L twin cpu system was ordered for a central system and data server, following examination of systems at CERN. At this stage the VMS service was performed by a cluster of an AXP3000 and AXP3500; these had quite extensive SCSI DECstorage farms for disk storage. These were later integrated partially into the newer systems.

The database for experimental results depended on DEC-Datatrieve on the VMS systems, and most data was stored in VAX11 binary. A decision was taken early to refer to a simpler filesystem approach to store these data, and to use ASCII files to avoid problems of binary/little-endian/big-endian compatibility. The basic format for ASCII data export since 1982 was the fixed length records of the TAPDAT format, which was easily extensible and was adopted, with a few moor additions for the Unix systems. To save space, and offer better subsequent network utilisation the ASCII data were compressed using an open source variant of the Unix compress tool on the VMS systems. There was an advantage on the VMS systems for using fixed length records; this allowed big files of time of flight data etc, to be read as direct access on VMS. For the Unix systems the huge file caches built in to them made this less necessary, especially given the faster processors. The fixed length format was always easy to read at a human level. There was an opportunity to add commentary zones on the contents of the metadata; this was done for many of instruments on the conversion of the instrument computers to Unix.

The conversion occured while the reactor was refurbished (1991-1995).  On restarting the reactor there were fewer instruments due to a reorganisation. Most still were using DEC computers, and the data conversion was performed on the central VMS cluster.
The staff departures, encouraged during the long shutdown, resulted in a number of scientists busy in software development for their own use being markedly diminished as they either retired early, or left for active science elsewhere. New replacements (in 1995) were not used to the “crude” third generation software tools and adopted packages, notably IDL, MatLab and Igor, which provided GUI interfaces and basic mathematical functionality. These proprietary systems were expensive, and initially the programs could not be exported easily, unlike the halcyon days of the VMS systems, where the executables could be shared and run anywhere. A great deal of time was spent on the GUI, rather than developing the underlying scientific analysis.

The rapid drop in price of workstations, especially after about 2000, when Linux started to be introduced on desktop PCs lead to the unheard of power of offering each scientist his own workstation. From a total computing capacity of some 500 MIPS on the Central AXP cluster in 1993 it was possible to envisage each scientist having similar computing power on their own desks. Some software systems were examined, where this unused power at night could be shared, but for the most part the type of computation done at the ILL were not so time consuming. Crystallographic problems which had run overnight in the 1970s could be completed within a few minutes on modern machines.

Typical RISC CPU speeds in 1995 were 300-500MHz. It is more difficult to compare with the VAX CISC instructions which rarely exceeded 20MHz! Graphical interfaces were capable of huge consumption of this resource; two X-window scrolling terminal windows which overlapped required considerable cpu power to evaluate whether it was necessary to update every pixel in the two windows. With the advances of graphics processors under Microsoft Windows there were even more attributes per pixel to consider, and again much of the processor was consumed in this activity, as was most programming effort. After the restart in 1995 instruments were changed to Unix using DEC OSF AXP-114 computers, then twin-cpu IBM-PCs running linux; electronics were addressed through ethernet links.

Back to main page       Top