[quicwg/base-drafts] Change max_packet_size to max_datagram_size (#3471)

[Kazuho Oku <[e-mail]>]

At the moment, the max_packet_size Transport Parameter is a size limit of "a packet," and we seem to interpret that it means "a QUIC packet." I am not sure if that has been our intent.

IIUC, the reason we have this TP is to tell the peer the maximum size of a "UDP datagram" that the endpoint can handle. Not the maximum size of a "QUIC packet."

The problem with having this limit being defined at QUIC packet-level is that it allows the sender to send a coalesced packet that consists of multiple QUIC packets, each of them consuming max_packet_size bytes.

For example, when responding to a client-sent Initial, a server is allowed to send a coalesced packet containing one Initial, one Handshake, one 1-RTT packet, each of them in full size, and expect the client to not drop the coalesced packet, assuming that the network can handle jumbo packets<sup>1</sup>.

This means that a QUIC stack needs to be prepared of handling UDP datagrams at most `3 * max_packet_size` bytes.

I do not think we would want to do that. Therefore, the proposal is to change the definition of the TP from max_packet_size to max_datagram_size.

As an alternative, we can change it to max_ip_packet_size, but considering the BSD socket API, I think that max_datagram_size would be a more natural fit.

[1] Note that there is a performance incentive to configure QUIC endpoints to send jumbo packets.

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/quicwg/base-drafts/issues/3471