RE: draft-montenegro-httpbis-h2ot-00

[Gabriel Montenegro <[e-mail]>]

Hi Emile, thanks for your comments.

A typical use of multicast is for either discovery or routing. If discovery, that would be handled via mDNS as is already done in some IoT scenarios, instead of overloading the application transport protocol to also fulfill that function. Routing also would use a different protocol, not HTTP/2. As for applications that require multicast, it is a bit complex, yes, as currently http/2 does not support multicast. There was some effort a long time ago to specify that (https://tools.ietf.org/html/draft-goland-http-udp) so one could consider going that path. However, I would caution against the security issues that raises (e.g., TLS won't handle multicast either even though there were some initial ideas that were being discussed in DICE WG for DTLS). 

Power consumption is difficult to assess. Mainstream protocols may be somewhat more inefficient than purpose-built protocols. The point is that if it’s a difference of 10 or 20%, that may not be relevant after some time to counteract advances in efficiency of implementations, or tradeoffs of knowledge, trained people, etc. Furthermore, there are examples of hw support appearing for mainstream protocols (e.g., for mdns) which would not be available for other discovery protocols.  This is another reason mainstream protocols may end up being more efficient.