Thread Links | Date Links | ||||
---|---|---|---|---|---|
Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
Also the source address and destination address must be available befoe
a forward decission can be made. I think that we should not be that picky
about what cut-through is. Everyone need to store some parts of the packet
(at least the header) to process the packet. The question really comes
to wether the delay depends on the packet size or not and wether it does
so regardless of of contention or not. Even cut through may store
a complete packet due to contention.
If a CRC is used to validate the header, it should only cover the header. There may be a seperate CRC to cover the payload. In my view a payload error should not lead to having the packet discardet, whereas a header error should defenitely discard the packet. We want to monitor payload CRCs as a part of performance monitoring and performance degradation location. The destination node should be able to monitor the same payload CRC error in order to collect _flow_ performance statistics. Robin Martin Brewer wrote:
--
|