- From: Ben Schwartz <bemasc@google.com>
- Date: Tue, 6 Feb 2018 11:18:19 -0500
- To: Göran Eriksson AP <goran.ap.eriksson@ericsson.com>
- Cc: Lucas Pardue <Lucas.Pardue@bbc.co.uk>, Martin Thomson <martin.thomson@gmail.com>, HTTP Working Group <ietf-http-wg@w3.org>
- Message-ID: <CAHbrMsB3L+3oHAZs_sow4cWGiQsMj8wSd0UMAoFZ6FAr9Hoqwg@mail.gmail.com>
On Mon, Feb 5, 2018 at 4:28 PM, Göran Eriksson AP <
goran.ap.eriksson@ericsson.com> wrote:
>
>
> On 2018-02-05, 18:55, "Lucas Pardue" <Lucas.Pardue@bbc.co.uk> wrote:
>
> Martin Thomson wrote:
>
> >I'm surprised that no one has mentioned TURN yet. So let me be the
> first.
>
> I'm not overly familiar with TURN. Are you saying that the
> capabilities of this new method are already fulfilled by TURN, that TURN
> could be enabled by this method, or that there are design elements to crib
> from TURN.
>
>
> Since Ben added "WebRTC" to the discussion, Martins comment make perfect
> sense. Personally, I would be waiting for Ben to detail the "WebRTC" case
> he has in mind, which would include TURN of course, (.
>
My feeling is:
1. It would be nice for HTTP/QUIC to be self-proxying, in the same way that
HTTP/2 and its predecessors are.
2. The obvious way to do this is to add a UDP proxy method to HTTP.
3. If we're going to define a UDP proxy method, it might as well support
the other common UDP-based protocols.
As for interaction with WebRTC, I would note that right now, running WebRTC
over an HTTP proxy greatly limits the use of UDP candidates, impairing
media quality and often causing suboptimal routing (through TURN/TCP). If
HTTP proxies could support UDP traffic (on the external interface), this
could be improved.
The main use case I had, HTTP/QUIC, does not suffer from peer
> communication issues so I find it hard to see where TURN comes in.
>
> +1. The HTTP/QUIC case is pretty straightforward (even though use cases
> are always useful of course). "WebRTC" will require an effort to clarify
> scope and proposed solution.
>
I agree, HTTP/QUIC seems like a simpler case. A "connection-oriented" UDP
proxy protocol (i.e. a single-destination tunnel) might suffice for
HTTP/QUIC, but it would not be convenient for WebRTC, because WebRTC (i.e.
ICE) works best when clients can receive packets from unexpected sources.
This has implications for the client<->proxy protocol.
Attachments
- application/pkcs7-signature attachment: S/MIME Cryptographic Signature
Received on Tuesday, 6 February 2018 16:19:21 UTC