[Server-devel] let's write a compiler!

[C. Scott Ananian]

On 7/10/07, David Woodhouse <dwmw2 at infradead.org> wrote:
> We're currently exploring the ways in which the server will act as a
> 'mesh portal' and provide IPv6 routing. Very little of the higher-level
> stuff has even been discussed, as far as I'm aware. The most useful
> thing you can do right now, if you're desperate for a server-related
> task, is either test all the software on the laptop to make sure it
> works correctly in an IPv6-only environment, or look at implementing
> something akin to NAT-PT which will allow TCP connections to Legacy IP
> addresses.

To expand on this short list of possible tasks:
 - I would really like to see an implementation of IPv6 SEND (RFC3971)
at some point, although this is not likely to be available for first
release.  I have not yet been able to locate an implementation.  Maybe
one exists, floating around one of the Linux IPv6 projects.  Locating
or writing one would be very valuable (and require very little
short-term coordination with the rest of the project, minimizing the
f2f needed).

 - At the moment, IPv4 connections are funneled through whatever NAT
the school has on their IPv4 connection.  This may restrict the # of
open IPv4 connections dramatically.  We'd like to do better.
NAT-PT/SIIT have been implemented by various hardware manufacturers,
but have serious problems (see
http://www.ietf.org/internet-drafts/draft-ietf-v6ops-natpt-to-historic-00.txt
).  Another option is and IPv4-to-IPv6 translator for websites akin to
http://www.sixxs.net/misc/coolstuff/ which could run on the school
server.  Since the school server already runs a (caching) proxy for
bandwidth conservation, a clever hack might be for the school server
to do something clever to serve IPv4 content.  (IPv4 connectivity is
assumed to be NATed and not globally routable, while IPv6 connectivity
is globally routable but the IPv6 tunnel endpoint may not be terribly
local.)

We hope to have a livecd build of the school server shortly (Dan Margo
is working on this), which might make it dramatically easier to work
on this stuff.
 --scott

-- 
                         ( http://cscott.net/ )