On IPG & Max packet size
This has come up before in logic track discussions, but the way I read the
max packet length, there is nothing in the standard for 10gig that says a
longer packet is 'not' legal. And there is no mechanism I am aware of that
prevents a longer packet from coming through to the mac.
The main issue of big packets is design of elastic buffer. As is to be
it is becoming tougher with wider data paths, lesser gaps (in terms of
continuous nature of stream (as opposed to idle IPG). In one of your earlier
you made a comment that multiple hops may be eating the IPG. All this makes
of EB very difficult. I propose following:
a. Make sure that at least
a1. two clock (32 bits) of IPG including SOP/ EOP OR
a2. one clock (32 bit) of IPG excluding SOP/EOP
is gauranteed at every node.
b. Decrease the clock tolerance level to 50 ppm (that reduces the EB
requirement to half).