Fixing a dead mainframe by listening to it over the phone
Way back in the early 80's I worked in the support centre for a large British Computer manufacturer.
I got a call from a site saying that the machine wouldn't load, and that they hadn't changed anything.
Although they did get a printer dump, it was particularly useless, as the only error it contained was KEVM_LAST_USER_OF_GLOBAL_EVENT, which meant that the main VM had failed, of course any stack frame revealing the cause of the error would have long been overwritten.
The processor they were using had a speaker which clicked on every CALL instruction resulting in various levels of noise mixed with gently descending tones. I asked them to reload the machine, and to put the telephone next to the speaker so I could hear the processor rebooting.
After hearing the load I gave them a list of load options, and the machine loaded. They thought I was a genius.
All I did was to notice that the load had almost completed before it failed, and I told them to switch off the last few load options.
Turned out to be a failure in the floating point unit, which caused the VM to crash, but the only option to use floating point in the load sequence was a hardware monitoring stats package.
Biting the hand that feeds IT
