Add a section with notes on PRE multicast replication behavior #513
Conversation
Also restrictions on what kinds of items can be added to multicast groups.
|
@samar-abdi I don't see how in Github to add you as a reviewer, but please take a look at this and add any comments you have. |
| (egress_port[N-1], instance[N-1]) | ||
|
|
||
| When a packet is sent to that group, `N` copies of the packet are | ||
| made. Copy number `i` that is sent to egress processing will have its |
There was a problem hiding this comment.
This is a very strong requirement. I do not think that the (port, instance) pairs should be assumed to be in any order and I would not require the replication to follow ay order either.
There was a problem hiding this comment.
I didn't read the statement as imposing an ordering, just the pairing. Each copy will be uniquely associated with a pair.
There was a problem hiding this comment.
I agree that this text could be misread to imply that copies are generated in a particular order, and I can try to come up with a different way to describe it that doesn't imply that.
In Section 6.2 of the current PSA draft, there is a requirement that PSA implementations process packets in the egress control block in the same relative order that they were processed in the ingress control block, for multicast packet copies with the same (ingress_port, multicast_group, class_of_service, egress_port) values (OK, it doesn't say it exactly that way, but perhaps it should). That has implications on how packet replication can be implemented. It doesn't imply that the packet replication set is a list with replication order determined by the control plane, but I expect many implementations would choose to do something like that.
If you think that shouldn't be a requirement, then we should discuss it, because otherwise you are probably going to make some multicast users pretty unhappy with your implementation.
There was a problem hiding this comment.
I've updated the PR with another commit intended to answer @vgurevich's comment.
Section 6.2 actually doesn't describe the FIFO properties as requirements, but 'expectations'. That may be wishy-washy in language, and we may want to change it to a requirement, or just a recommendation.
There was a problem hiding this comment.
@vgurevich Do these modified changes look better to you?
There was a problem hiding this comment.
Sorry for the late reply. Yes, the changes look good. Thank you!
Also hopefully clarify the expectation of packet ordering in a PSA system for multicast packets, that is earlier in Section 6.2.
Also restrictions on what kinds of items can be added to multicast
groups.