Re: [rtcweb] Non-media data service consensus and requirements

["Timothy B. Terriberry" <[e-mail]>]

> No, most likely it was I who missed something...  Yes, I remember the
> timer questions, but I thought Jonathan answered this.

IIRC, Harald suggested that the best way to decide if this was actually 
feasible was with "running code", at least at the proof-of-concept 
level. This seemed like a very good idea to me. The resolution of JS 
timers is not the only issue. One also has to deal with other pages with 
long-running JS scripts that prevent your timers from firing. Various 
advertisers do this on a semi-regular basis, though I don't have good 
numbers on how frequent this is, or how much of a problem it would 
really pose. Running code could be used to actually measure the 
connectivity success rate, for those that want to know.

I don't think the compatibility concerns have been addressed, either. 
With ICE in the browser, you have four or five implementations total 
that have to interoperate, vs. (potentially) one for each website. 
Updates to browsers can be deployed, at scale, to hundreds of millions 
of users in the span of a few weeks (at least with Firefox and Chrome, 
Safari and IE also have well-established short-term update mechanisms, 
though they're usually for security issues rather than new features), 
vs. trying to get each and every website to update their JS library. An 
individual website may be able to deploy updates more quickly than the 
browser vendors, but interoperability isn't about the individual.

And now there are the security concerns with sending datagrams after 
trying to short-cut the ICE process. I don't know if mandating STUN 
requests before sending data to a remote peer is sufficient to address 
these, but clearly some others see a potential problem. Personally, I 
think having p2p datagram support is a much more important feature to 
end users than being able to change the connection handshake protocol in JS.

Everything above is an argument I've heard someone else raise at some 
point. I haven't heard a lot of good responses to these arguments.