Re: [tsvwg] The List (of application-layer desired features)

On Sep 5, 2013, at 4:05 AM, William Chan (陈智昌) <willchan@chromium.org> wrote:

> I think it's a fine thing to support long-term solutions that aren't immediately deployable, assuming of course that's the right long-term approach. But I think it's a mistake not to deploy something in the meanwhile. Let's not let the perfect be the enemy of the good.
> 
> For example, I think it's great that SCTP can work over UDP. People may argue that it should really be deployed in kernels, and it's probably true that it'd perform better that way, but it's significantly more deployable when you layer it on top of UDP. So I think it was a smart choice to deploy rtcweb using SCTP over UDP, rather than arguing that we should wait for SCTP/IP to be deployed in relevant OSes and NATs/intermediaries.
Just to be clear: You can implement SCTP/IP and SCTP/UDP/IP either in the kernel or
in the application. For SCTP/IP some OSes require specific privileges, most NAT boxes
don't support SCTP/IP. So choosing something like SCTP/UDP/IP allows for deployment
within the application and make use of kernel implementations when available.
Therefore, the crucial point is middlebox support which requires the UDP encapsulation.

Best regards
Michael
> 
> 
> On Wed, Sep 4, 2013 at 6:05 PM, Joe Touch <touch@isi.edu> wrote:
> 
> 
> On 9/4/2013 5:55 PM, William Chan (陈智昌) wrote:
> ...
> 
>     My point about TCP-AO is that it's not useful to continue to
>     complain that something isn't widely deployed. If you need what it
>     has, then why not start by helping deploy it? Then at least there's
>     one less thing to complain about being missing in a few years.
> 
> 
> So I actually don't think my browser needs TCP-AO, so that's probably a
> bad example.
> 
> Sure.
> 
> 
> But in any case, I don't know what I can do to help get it
> deployed. Are you suggesting I periodically email Microsoft to ask them
> to update their kernel networking stack to support it in their OS? What
> help would you like from me here?
> 
> We need those who do need it in endsystems to either contribute code or support others to code it.
> 
> There was a long discussion in RPKI about the fact that router vendors want it in host OSes (to run as RPKI servers), and who support it in their routers. They spent a lot of time complaining that it wasn't available in the host OSes, and a lot of time considering alternatives, rather than either coding it up or supporting someone (anyone) who could.
> 
> That has happened before with other extensions. It's not useful to claim "it's not out there", especially when such a solution better or more generally useful. I.e., claiming that "it has to already be deployed" is a silly pre-requisite for *this* community; if it's that ephemeral a problem, IMO it's not usefully solved here.
> 
> Joe
>

Received on Thursday, 5 September 2013 05:09:45 UTC