Re: [rtcweb] A plea for simplicity, marketability - and... who are we designing RTCWEB for?

[Wolfgang Beck <[e-mail]>]

Comments inline

On Fri, Oct 21, 2011 at 1:47 PM, Ravindran Parthasarathi
<[e-mail]> wrote:
> ...
> Please note that SIP is not designed in legacy
> telephony (PSTN/SS7) architecture fashion  (by ITU) but designed based on
> Internet model wherein UA(endpoint) is intelligent.
..but we still introduced the BYE request and stateful proxies for
forking.  Both functions could have been done
in the UA..

> After 10 years, now we are interested in real-time communication using HTTP only.
No, not at all. We're interested in UAs that are programmable on the
fly -- like a web browser with javascript --,
so that we don't have to specify a protocol between client and server.
 Jonathan Rosenberg did an
enlightening presentation at one of the past plenaries.

When do we need a standardized protocol? We need one if two parties on
the same network layer were implemented
by different vendors.

This is true for RTP (at least if we stick to the P2P media dogma).
Browsers of different vendors must interoperate.

It might be true for SDP's functionality. If we implement it deep in
the browser, it must be standardized. The problem with SDP is, that it
relies on some other protocol for transport and has some very specific
requirements on that protocol. If we can make sure that all parties
involved in a call use the same JS client, we can get away without
standardization.

At the layer of call signaling messages are ultimately exchanged
between two JS clients running on browsers. If we have different JS
clients here, we need a standardized protocol between them. With
trapezoid interconnection, we can hide this to some extent by doing
protocol translations between JS client A / RTCWEB Server A and Server
B / JS client B. Some
information will get lost in translation.

But as browsers can load JS on the fly,  user A could simply point her
browser to Server B and use the same client as user B. Now all
components at the signaling layer are provided by a single vendor and
no standardized protocol is required. There is no loss of information
as there are no protocol translators.


Wolfgang Beck