IBM Corp’s plan to save the mainframe appears to be founded on creating a single-chip CMOS uniprocessor with the performance of a mid-range 9221 – that’s a high-end 9370, and creating a highly parallel machine that will include up to 48 of these. They will be organised as up to eight tightly-coupled six-processor nodes, linked together using IBM’s Sysplex clustering facility to create a machine that should deliver performance a little over that of the ES/9021 Model 822, which is the two-processor model in the water-cooled ES/9000 line. IBM reckons that this machine will be dramatically cheaper to build, so that it will be able to offer today’s performance at as little as a tenth today’s price. The machine will look neat enough, with the eight nodes tucked tidily into a compact boax, but the devil of course lies in the detail. The detail in this case is the creation of versions of the key system software to run efficiently on such a configuration. There will be so-called multisystem releases of all the old favourites, with a read-only version of DB2 the first up. That will be followed by a read-write version, a multisystem version of CICS, improved versions of MVS and IMS, and a Job Entry System workflow manager. A central workload manager will attempt to minimise bottlenecks and will assign jobs to specific processor nodes. The learning curve will be tough, it is by no means clear that the concept will be an easy sell to today’s risk-averse MVS user, but IBM’s immediate target is a 40% a year reduction in mainframe manufacturing costs. The global mainframe slump has caused a deal of pain at IBM already and there is more to come with the transition to CMOS. By the end of this year, IBM says it will have cut employees and plants associated with mainframes by 40%, primarily at its three locations in the Hudson Valley, New York. In July, the mainframe work force stood at 13,800 employees, down from 26,400 in 1990 and a peak 31,306 in 1984; more cuts are expected in Europe this year.
