winscape 2347

for a little topic drift ... there was an industry service that collected erep data (ras and error recording data) from large number of customer installations .... somewhat removed customer identifying information and published summaries on machines. customers could compare reliability data on different machines, models, disks, etc. both from vendor as well as all the clone processors, controllers, devices, etc.

in late 1980 time-frame, stl was going to move something like 300 ims programmers to an offsite location. they really hated the idea of having to access systems at stl via remote 3270. I wrote HYPERchannel driver support that was basically a form of channel extension ... which allowed moving "local" kind of devices to remove location ... and sort of allowed them to operate over high-speed telco links.

as part of doing the driver support for channel extension ... i had to translate various kinds of unrecoverable telco errors back into some form of emulated local channel error (where then the standard ras support would do additional retries-restarts as well as record the information).

when 3090 processor had been in customer shops for a year ... one of the 3090 product managers tracked me down. they had predicted that for a certain clbutt of channel errors, the 3090 would have approx. 3-5 such errors .... per year across ALL 3090s (i.e. not 3-5 errors per 3090 per year ... but 3-5 errors per year total aggregate across all 3090s). the industry service was reporting something like 15-20 total errors of this kind for the previous year across all installed 3090s.

it turned out that the emulated channel error condition that i had chosen had gotten out into driver support that was shipped to some customers ... and that the additional 10-17 errors for the 3090 product line for the previous year was coming from a customer that had this channel extension software installed. the report of these additional dozen or so errors for the 3090 product line year was causing an enormous commotion.

so i looked at the problem for awhile ... and there was another channel related error condition that i could emulate ... that for all intent resulted in the same RAS process ... but didn't inflate the statistic of specific interest.

a couple recent postings on the same subject Thou shalt have no

misc. past collected hsdt postings

previous post in this thread mentioning time-sharing and 3274

other recent postings touching on time-sharing subject parallel x86 design parallel x86 design parallel x86 design parallel x86 design parallel x86 design