An ISP I was working for in the late 1990s, early 2000s were looking at acquiring new modem kit with higher port density (yes, this was totally sensible at the time). Among the providers we were looking at were Nortel.
Now, we had a price separation between multi-link PPP and single-link PPP and for a variety of reasons, we did this by having the multi-link calls terminate in dedicated equipment.
This worked fine for a day or so, until the Nortel CVX started rebooting in "peak dial-in time" (somewhere in the 3 PM - 4 PM bracket). And we could not for the LIFE of us understand why. Got Nortel on the case, they could not even recreate the crash in their lab. Until I happened to ask about their simulation parameters. They were doing a 1 multi to 5 non-multi mix, at about 75% of the incoming call rate we had.
That, alas, was not enough to tickle the bug, which was basically that the CVX had to shift either the first or the second call of a pair onto the same DSP, and if a new cal arrived between "start the move" and "finish the move", the call being moved would simply be lost to data corruption. And when the watchdog would go through and check for internal data sanity (once per second or so), it'd fall off the data chain and the watchdog timer would reset the chassis.
Nortel declined to fix the bug. We declined to keep (or pay) for the equipment. I was not massively surprised with Nortel eventually folding.
Now, we had a price separation between multi-link PPP and single-link PPP and for a variety of reasons, we did this by having the multi-link calls terminate in dedicated equipment.
This worked fine for a day or so, until the Nortel CVX started rebooting in "peak dial-in time" (somewhere in the 3 PM - 4 PM bracket). And we could not for the LIFE of us understand why. Got Nortel on the case, they could not even recreate the crash in their lab. Until I happened to ask about their simulation parameters. They were doing a 1 multi to 5 non-multi mix, at about 75% of the incoming call rate we had.
That, alas, was not enough to tickle the bug, which was basically that the CVX had to shift either the first or the second call of a pair onto the same DSP, and if a new cal arrived between "start the move" and "finish the move", the call being moved would simply be lost to data corruption. And when the watchdog would go through and check for internal data sanity (once per second or so), it'd fall off the data chain and the watchdog timer would reset the chassis.
Nortel declined to fix the bug. We declined to keep (or pay) for the equipment. I was not massively surprised with Nortel eventually folding.