Re: [rtcweb] Isolating data channels (Re: Asking TLS for help with media isolation)

[Harald Alvestrand <[e-mail]>]

On 04/07/2014 07:01 PM, Martin Thomson wrote:
> The subject is largely speculative, but when it comes to solutions,
> the problem will not be a lack of options, but an inability to choose
> a "right" one.
>
> On 7 April 2014 01:02, Harald Alvestrand <[e-mail]> wrote:
>> Wild suggestion: if you want per-track isolation properties, open up a data
>> channel with a protocol called '*WebRTCIsolationInfo' and use it to send
>> information about the isolation status of each track, thereby also providing
>> a working example for the rule 'all data channels that have protocols
>> starting with "*" are for browser internal usage'.....
> I assume that you are talking PPID here.  Given that it's off limits
> for JavaScript currently, then it does provide an opportunity for this
> communication.  The problem there is that you need to spin up data
> channels, even if the application has no need of them.  That's a
> fairly high cost.
I was actually thinking "protocol" as in the string that goes into the 
datachannel setup packets. PPIDs would work too for separating 
browser-to-brower from app-to-app, but I wasn't thinking of them.

Yes, data channels do cost something to set up. But we're already paying 
the DTLS tax in order to set up the keying, so it's "just" another 
request/response. How many extra round trips does the SCTP setup add?

(Would need the response in order to make sure the respondent gave the 
promise to obey isolation, I think.)

(and to Matthew: At least we wouldn't have *yet* another congestion 
context to manage, which would be the case with a separate TCP 
connection. There are always tradeoffs.)