V6 Ethernet Overflow

[neelay893]

ROACH112 would consistently have ethernet buffer overflows while sending photon data, usually within the first few minutes of starting packet stream. Both gbe64_tx_afull and gbe64_overrun were set. Can be fixed temporarily by resetting the buffer. This is the first time we've seen this happen, and the design of the block hasn't changed in several months.

[neelay893]

Looked at some of the data from when certain feedlines were cutting out. It does seem like short periods of very high count rates are causing buffer overflows. On the two boards that cut out (FL8a and 8b), all pixels had ~10 counts in the 0.5 ms before the board cut out (which is 20,000 cps!). Couple of odd things I noticed:

• The flashes on each board were ~1ms apart; maybe the PPS isn't synced perfectly between boards?
• Only saw flashes on FL8 (in the +/- few ms range from when the boards cut out)

File used: dark:/mnt/data0/neelay/ProcessedData/Subaru/20181222/1545553145.h5