Authors, The multi-stream procedures in -03 are much improved. Thanks.
I'd recommend that you always establish the control stream. This avoids ambiguity about stream zero. In the case of BGP over QUIC, the motivation for different frame types is that the control channel has need to be the back-channel for unidirectional function channels in some cases. What's the motivation for the sequence number? For statistics, while there's text suggesting that per-afi/safi statistics might be sent on the per-afi/safi streams, I'd suggest that these stay on the control channel. The motivation for this is trying to get better atomicity of the BMP-wide statistics vs. the per-afi/safi ones. When split, you have a greater chance to confuse things like total message counters and the more granular ones in time. I outright admit that there's nothing in the protocol that requires that statistics be gathered in atomic snapshots. However, there's not a lot of motivation for applications that can do so to not do it. But even such atomic state snapshots can lose temporal coherency if you serialize them over different streams where the messages show up at differenet times depending on queues. -- Jeff _______________________________________________ GROW mailing list -- [email protected] To unsubscribe send an email to [email protected]
