... when the janitor polishes the datacenter floor with the big electric floor buffer.
OK, I'm not seeing backup tape erasures. It's still that packet error thing I've been chasing for the past couple of weeks.
I'd been running tests with traffic sped up to 10× the nominal packet rate, to reduce the time-to-error.
When I return to the nominal rate? I get about 1/10 as many packets through, on average, before the error.
So it's not a question of bit error rate, nor packet error rate, but something related to time outside of the message formatting / encoding / transmission.
Kind of like the classic "the circuit glitches when someone uses the nuker in the break room" scenario. Except that this has to be internal to the system, given that it (or something very similar) is happening both there and at the client site.
Grrrr. Currently testing with a polling routine slowed down, despite the fact that nothing the routine is currently doing should have any effect on communication.
Comments