eugene@pioneer.arpa (Eugene N. Miya) (03/09/88)
Marc Gibian writes: >Many of the CDC Cyber machines of the high side of the 170 series used banked >memories, along with multiple functional units. Once in a while you would run >into an algorithm that would get into phase with itself and try to make all >memory accesses from the same bank, and would be mysteriously slow. These >were -real FUN- to fix. Antecdotal (sic) information such as this keeps computer performance measurement a black art. We have to take the "mystery" out of things like this so we can get on with more real fun things. More people have to publish their pathological cases of code; we may find they are more pathological on more than single architectures. When we get the combined architects on the net working to solve problems, then the net will have advanced the state of the art, but until then..... From the Rock of Ages Home for Retired Hackers: --eugene miya, NASA Ames Research Center, eugene@ames-aurora.ARPA "You trust the `reply' command with all those different mailers out there?" "Send mail, avoid follow-ups. If enough, I'll summarize." {uunet,hplabs,hao,ihnp4,decwrl,allegra,tektronix}!ames!aurora!eugene