DOS360: Forty years 532

to some extent it was transfered bytes per mip ... and-or arm accesses per mip.

in the late 70s ... i started making some observation that disk relative system performance had declined by something like a factor of ten times (by the early 80s, cpus, memories, etc had increased by a factor of 50 times while disk arm performance had only increased by a factor of five times or less ... therefor the disk arm performance had a relative system performance decline of a factor of ten times).

we had a 360-67 that support 70-80 cp67-cms users ... and much later there was a processor with nearly fifty times the processing power supporting 300 vm370-cms users. It turns out the disk arm performance had possibly improved by a factor of four times.

One of the things you really started seeing was trying to leverage various kinds of caching (explosion in the availability of memory sizes) to compensate for the lack of disk performance improvements. Other approaches that tended to trade-off electronic memory vis-a-vis disk arm performance was to transfer in significantly larger blocks (which could make sense for some application environments).

In any case, my buttertions upset some number of people in the disk division ... and their performance modeling group got buttigned to refute it. After a couple months, they came back and effectively said that I somewhat understated the decline in disk relative system thruput. They then turned the study into SHARE presentation ... where they described disk strategies to help overall system thruput.

some random past postings mentioning the decline in disk relative system performance:

--