on a control packet handling..
Folks,
Expanding one offline discussion to a wider audience.. we have our
current packet & header format for an RoE control packet. That's fine.
However, there are different uses for the control packets i.e., some are
truly for "control" and most likely always ending up to a CPU to process
with request-reply/ack semantics etc. Then we got data related control
packets e.g., originated by a CPRI mapper and these packets are likely
to be processed in the data path or the mapper's "control process" as
the draft calls it. These control packets have no acks and they are
uni-directional etc. Now the question is that whether we should indicate
this "early" in the RoE header? It is possible to do this by looking
into the RoE control packet opCode field (used to be the "subType") but
it might make sense to be able to make the decision already by examing
just the subType field (used to be the "pktType").
Examing just the subType (was "pktType") would call for two classes of
RoE control packets. Any comments/thoughts?
- Jouni