RoE header and mapper thoughts: number of AxC per packet
Folks,
We touched the issue of the number of AxC per packet during the f2f meeting and at the end of the last call. Obviously for the structure agnostic mode, we do not care, since the mapper or encap/decap does not care about the content structure/format it transports.
For the structure aware mode, I think the possibility to switch at the individual AxC flow level has great value. One could, e.g. have a config where one RE feeds different antennas to REC_1 and REC_2 and so on. Allows great deal of flexibility for the system design and makes switching easy.
Therefore, I would propose, we assume one AxC per packet. If there are multiple AxC flows between same endpoints (say RE_1 and REC_1 or one SrcAddr/DstAddr pair), we need some identifier to differentiate between the AxC flows. The identifier is something that is not needed by switching/transport functions, only the mapper & encap/decap functions who need to look into the RoE payload. In my RoE header proposal I had 7 bits reserved for AxC identifiers. Currently that is "too large". However, for the future massive MIMO that might be too small to begin with.. unless we can assume the underlying transport or system design is able to provide additional tools for partitioning (like VLANs or using more than one REC etc).
One more thing. The above identifier description assumes the AxC identifier number space is unique only between two endpoints. Endpoints in Ethernet case would be defined by a DstAddr & SrcAddr pair.
Any comments?
- Jouni
--
Jouni Korhonen, Ph.D, Associate Technical Director
CTO Office, Networking, Broadcom Corporation
O: +1-408-922-8135, M: +1-408-391-7160