[1510] in Commercialization & Privatization of the Internet
Alternate Routing
daemon@ATHENA.MIT.EDU (Eliot)
Fri Oct 18 14:05:36 1991
Date: Fri, 18 Oct 91 11:04:28 PDT
From: Eliot <lear@turbo.bio.net>
To: com-priv@uu.psi.com
At the com-priv BoF at Interop I had an opportunity to sit down with
Yakov Rekhter and discuss the routing problems of fall over on a
regional when there is a commercial provider involved. This is the
problem I brought up some time ago on com-priv. He and I put together
this message, as a result of our discussion. Here (again) is the
problem:
Given a regional that wishes to accomodate academics connecting to
ANS/NSFNET and commercial sites connecting to CIX, establish a routing
mechanism that enables the sharing infrastructure such that fallover
not cause naughty bits to traverse the NSFNET. The assumption is that
traffic from a specific network must remain of the same flavor
(commercial or R&E) - that is, the granularity of this method is at
the network number, not at the packet (thus putting aside Steve
Knowles' points for the moment).
Solution #5 (as I recall, suggested by Yakov):
Assuming that the CIX router is on the same subnet as the ANS router,
each feed router shall keep a table of which networks are commercial.
It is assumed that there will be a centralized registry that would
provide information about whether a particular network is commercial
or not. Each feed router retains routing information it receives from
both CIX and ANS. When a feed router is presented with a packet going
out of the regional, the router checks source and destination IP
address of the packet. If both of them are commercial, then use
routes provided by CIX. In all other cases it is up to the regional
to decide whether to use CIX or ANS (C-R, R-R traffic).
This scheme requires the following:
1. A table that lists commercial networks.
2. Two routing tables within each feed router (C-C allowed, C-C disallowed)
on the shared Ethernet (where CIX and ANS routers sit)
3. Ability to select one of these routing tables based on commercial
networks table lookup.
You may think about all this as an extension of QoS based routing,
where QoS is used to mark C-C traffic. QoS in this case is not present
as IP QoS parameters, but is derived from the source and destination
IP addresses of the packet. You may call this QoS "C-C".
A packet has an implicit "C-C" QoS iff both source and destination
IP networks are labeled as commercial. A router maintains two
forwarding tables: one for "C-C", and another for the rest.
It'll work, but here is the bad:
As the distance between the ANS/NSFNET router and the CIX
router increases, more routers need to be involved.
You have to roll your own router.
You have to maintain the table of Commercial sites on every
router that participates in this game.
The cost is essentially an extra routing table in each router,
plus a lookup on both source and destination addresses per
packet.
The good:
You can get away with using existing routing and network layer
protocols to communicate.
Comments?
