Taking a step back from the detailed discussion, it seems we don't have
any major objections from the UK community to the proposals for
deploying this I circulated last week, so I feel happy to take them to the
RIPE DNS WG to take them a step further.
We have offers of resources from VBCnet and NOMINET in the form
of server hardware and operational support, and various LINX members
as regards transit. We have support in principle from LINX, ISPA
& NOMINET.
Here's what I think the outstanding issues are:
- fitting it all in to the European and Global picture:
- for further discussion
- routing policy for the NS ? various points here:
- do we put the server in the LINX AS, or create a new one ?
For this point it is worth remembering the reason we have the
LINX AS in the first place, namely:
1. To make the LINX LAN prefix visible to the rest of the world
2. To give the LINX secretariat Internet connectivity
3. To have something everyone can peer with to gather
useful stats in the collector.
I don't think there are any sacred cows for any of these here -
if we adjust the LINX policy to meet requirements of the root NS
rather than the other way around, I don't think we break either
of 1 or 2. I don't think it breaks 3. either, but this requirement
will go away when we have a route server in any case.
i.e. I am very appreciative of the existing LINX members who
give transit to the LINX, but, it would not break my heart to have
the same connectivity/policy for the LINX as a UK-based root
NS, and would understand if the existing providers did not
want to fully transit a root servers' traffic - we have had enough
offers to transit the root NS I do not see this as a problem.
- how many members should provide transit to the root NS AS ?
we have appear to have 2 schools of thought here, namely
that either everyone should transit the thing to spread the load
as evenly as possible, or that we should only have a
small number to keep the routing deterministic (there's also
a 3rd more wacky suggestion to do some special local
hack, but I tend to agree with Paul Vixie's comments on that).
My view is that we have to be careful about too many members
providing transit, as we can easily finish up with lots of views
of the way to the LINX AS with equal path lengths, but
non-obvious tie-breaks, leading to hard to predict connectivity.
Also increasing the number of path/prefix products in the global
routing tables is in any case ecologically unsound.
Perhaps the solution is for a small number of well-connected
LINX members to provide universal transit, but for an additional
group to provide additional transit to specific places (like the
Nacamar offer).
(Further discussion of server routing policy is probably best
confined to technical forums, i.e. maybe nom-dir and ispa-com
want to be dropped off the distribution.)
- server for just "." or TLDs ?
This kind of ties in with routing policy, in that those providing
transit have potential concerns about traffic levels generated
by a TLD server . One solution here might be PeterL's
suggestion
of seperate servers for TLD and ".", with different members
providing different transit to each. The TLD server does not
need to have its own independent address space in the same
way the "." one does. I think we can still get away
with having them in the same AS, though there is some risk
of asymmetry here.
All we need is an additional offer of server hardware.....
- router hardware
We do need to think about upgrading the collector to a 64Mb-
capable box, most likely a 7200.
If we think that a 100Mb connection for the server is needed,
then the is a spare slot in the collector for an FDDI or 100baseT
card.
I'm sure there was another point to make, but being in the RIPE terminal
room is distracting me a bit. More discussion of this here tomorrow.
Keith