TOD Clock the same as the BIOS clock in PCs 3266

Shmuel Metz , Seymour J.

somewhat idle observation about memory contention ... but the 360-65 and 360-67 were essentially the same hardware with dynamic address translation (DAT) box added to the 360-67. The 360-67 could be run with plain vanilla os-360 operating system (w-o activating DAT) and it appeared to act in all respects like a 360-65.

the 360-67 multiprocessor was a different animal. it had provisions for multi-ported memory (processors and i-o access concurrently to memory). It also had a "channel controller" ... allowing all processors to access all channels ... something not seen again until 3081 with 370-xa.

the multi-ported hardware introduced about a 10percent additional memory access latency (but reduced memory contention). a 360-67 "half-duplex" running purely compute bound application ... had a "MIP" rate that was about ten percent less than a "simplex" 360-67 (or 360-65).

However, cambridge science center

had an opportunity to replace a 360-67 "simplex" with a 360-67 "half-duplex" ... running the identical same software, workload, and configuration. The "half-duplex" had approximately 20 precent higher processor thruput than the "simplex". This would appear to imply that on a standard 360 (65 or 67) that I-O memory contention could have possibly 30 percent impact on processor thruput (aka the "half-duplex" with no memory contention was nominally ten percent slower than simplex ... but under normal i-o load and memory contention, memory contention appeared to have slowed down a "simplex" processor until it was approx. 20% slower than "half-duplex" ... rather than 10% faster).