[consume-routing, old!] Current routing issues

[Stuart Henderson]

This is so old I hate replying to the list but I didn't see any more
about it, and well, I managed to pin down a few ideas I found running
around in my head, so here's as good a place as any, and it saves me
from having to think up a new Subject line, always a tricky problem (:

> People have said we can obtain IP address space - details needed of the
> process for this. Is anyone offering spare space?

Spare space within someone else's netblock doesn't allow separate
route announcements. (Well, the protocols allow it, but large ISPs will
filter these out - the global routing table is big enough as it is :)

If having consume on internet address space is a requirement, then
what is needed is some space from RIPE and maybe also to become an LIR.
Probably PA space (see http://www.ripe.net/ripe/docs/pi-pa.html for the
differences between PI and PA) and then that is sub-allocated to users.
However, you can't do much with this unless you can announce it on the
internet: it's going to be necessary to run a router somewhere which is
capable of announcing consume into the global routing tables.

This could be a PC (gated, mrt, zebra) or a hardware router, hosted
at a friendly ISP or at somewhere such as the LINX, LonAP etc. It can
announce routes to the world and maintain a stack of tunnels so that
it can forward packets onto edge routers of the wireless network at
whichever address it knows for them. These edge routers could connect
to pretty much any ISP via ADSL, Cable, SDSL over EPS8/9, nailed-up
ISDN, whatever.

Of course, seeing as this as near as possible to a perfect solution,
there has to be a big problem. You pay for your bandwidth twice :(
(Once from the ADSL/cable/... ISP, and once for whoever you buy
transit from, since I presume at some point someone's going to want
to get to a site connected by a provider who won't peer). Actually
if you're at the LINX there is a lot that you'd probably get from
peering (not sure how good an idea it is to mention some of the
names here, but take a look at the peering matrix on linx' website
and you'll soon get a pretty good idea who will and who won't!)
But it's definitely not the whole world. And this centralised
paying-for-bandwidth (even cheap bandwidth with low delivery
charges) sounds like it might be a problem for something running
on noncommercial spectrum[1]. And, ermmm, the LINX does cost
quite a lot... LonAP might well be worthwhile though - and it
does give a place to run multihomed without paying commercial
providers (so, it's maybe an easier way to obtain address space
from RIPE - and note, you don't necessarily have to have routes
to and from the whole world in order to obtain an ASN and address
space ;). Website is at http://www.lonap.net/ - it's physically
in Telehouse - prominent members include the BBC, Nildram and
Above.net. Though most of those would be more likely to be
hosting than initiating connections, which is kind of covered
by the next paragraph anyway.

For things such as accessing the majority of internet web sites
from a wireless node, it's not such a problem: nobody really cares
what address is on the packets so they can be NATted or proxied
as necessary (and there's plenty of ideas about catering for that:
freebsd and linux can both load-balance tcp connections nicely,
see for example ja.net's web caches).

And even for email and hosting web sites on consume, this can be done
just using consumer connections (though static IP is very preferable).
Run split-dns and gateway machines. Any mailer will handle the mail
(itself it would refer to internal consume dns, the world would see
MX records pointing to the gateways). Web hosting can be handled much
the same, using squid in 'web accelerator' mode. This is all there
now - the software is mature, it's all pretty straightforward to
configure, and it can keep all of the infrastructure within consume.

I wonder if IPv6 may be easier to have real addresses: freenet6.net
is pretty well-known, though I suspect there might be too much hackery
to have a few machines inside a single prefix connected via different
tunnels (seeing as it's running on someone else's machines), perhaps
uk6x.com might be a better place for experimentation...

-Stu
<stu at spacehopper.org>

[1] (hmmm, I always have wondered: does that mean it's not actually
okay for a business to use 802.11 for their own internal traffic?)