Discussion of large systems

On 1 Nov 2006 at 16:44, Billy Pettit wrote: ...Except when it wasn't. If you had a large enough bank of tape drives, the white noise from the vacuum pumps was pretty rough on the nerves after a few hours. I recall working on a system that had 4 Cyber 74's hooked through a MAC to 4MW of ECS. Part of the acceptance test suite was running a full load on the system and reading long-record stranger tapes. (That's the 1LT PPU overlay). Since a PPU only had 4K of 12-bit words, and the theoretical maximum size of a tape block was much more than this, you had to read 5 PP words at a time, then CWD the lot to CM. But ECS accesses had priority over PP access to the read pyramid. So, you'd start to read a tape block, note the "lost data" flag, backspace to the beginning of the record and try again. It was pitiful to watch. There was a feature called "Priority Access" where a PP could muscle its way into the pyramid by setting bit 17 of the CM address (this was before the 131K machines and the problems that THAT created). Unfortunately, this aborted any ECS transfers from the CPU, which would promptly restart them. Things got very slow--so slow that DSD couldn't even refresh the operator's display on a timely basis, so the screen got all dim and flickery. Management gave that solution thumbs-down. I never did figure out a good way around it and I think that 1LT remained a cursed overlay until CDC dropped the Cyber architecture. Cheers, Chuck