Here was an IBMer from the U.K. talking about cache controllers and the future of cache at IBM. He was a great speaker, and knew his stuff.

Cache controllers came about as a result of IBM having so much data on one drive. It is becoming more and more difficult for DASD managers to tune their I/O subsystems, so cache devices ease the load a little bit.

He mentioned that the U.S. and U.K. both had substantial appetites for disk storage, but that the two countries managed their DASD differently. In the days of the 3330 - the last removable packs IBM marketed - U.S. installations averaged 1.5 packs for every drive on the computer room floor. U.K. computer rooms had an average of 9 packs per drive! He joked that you could always tell who the English computer operator was - he was the one standing over there scratching his ankles.

The speaker discussed some of the technical problems involved in DASD system design:

• You can't increase disk rotation because of a friction problem.
• You can't do full track staging because you run up the device busy time.
• You can't generally use all of the cache memory because partial track staging causes fragmentation in the cache.
• You can't improve the response time for a cache hit substantially using current channel protocols.

• LARGE cache memories are coming. They may be merged with the 3090 extended storage (which goes to 2 gigabytes these days).
• Non-volatile cache memories are coming, probably via some battery backup scheme in the controller. This is the only way IBM can beat the requirement to write-through the cache on every write.
• EXPECT extensive use of FIBEROPTICS, and a re-work of the IBM bus-and-tag channel architecture.