AaronL writes:
On 08 Aug 2012 19:07:06 GMT, (Scott Lurndal)
wrote:
In this case it was past midnight, testing a completely new firmware and OS
on a working customer mainframe and the FE had gone home hours ago. Given the failure
was in a Load Index Register instruction, which was handled by the Fetch
module, I tried to swap the two (20"x30") fetch cards with the on-site spares
kit. The new OS still faulted. Tried the two Execute Module cards. Still faulted.
Eventually got around to swapping out the MCW (Memory Controller Write) card which
fixed the problem. All night, alone in the datacenter of one of Belgiums larger
banks. Probably wouldn't happen nowadays.
In the middle 60s I worked testing newly manufactured GE mainframes.
You would have thought board testing would have caught all board
problems but often when a machine had a problem it was caused by a
(new) bad board. So board swapping was a still valid troubleshooting
method even when testing at the new system level.
Company I co-founded in 2004 built an ASIC (distributed node controller). A
very complicated part. When we got the first (untested) samples from the fab,
we plugged one into a test board and ... nothing. I would have spent hours
trying to figure out why it wasn't working, but the ASIC architect just swapped
it for the next one in the tray and it came right up. We never did find a bug in
the part that required a re-spin or metal layer change by the fab.
scott