Hi Bill,

I will write in a separate message about fine-tuning your page:

  http://bill.herrin.us/network/rrgarchitectures.html  (Fifth Draft)

to properly cover Ivip.

Here are some thoughts on your Item 1: "Routing table size problem,
root cause:".  This explanation seems to assume a lot of knowledge of
why multihoming, TE and portability in the current system creates an
undesirable number of BGP advertised prefixes.

I think that the real problem is that with current techniques, there
is a serious mismatch between the number (millions or in the future,
perhaps hundreds of millions or billions) of end-user networks which
want multihoming, TE and/or portability and the only way this can
currently be provided - by globally routed prefixes for every such
end-user network, which every DFZ router needs to handle in its RIB
and generally in its FIB.

It is not open to us to say the problem lies in millions of end-user
networks wanting multihoming, TE and/or portability.  So the problem
is that the current architecture can't scale to meet this demand.

That is the guts of the problem.  It is only one interpretation of
the problem to see it in terms of IP addresses being used for too
many purposes all at once, which is your page's current definition of
the root cause of the scaling problem.

There are potential host-based solutions - with new protocols and
changes to stacks and applications - which pushes the responsibility
for multihoming, TE and portability to each host, by way of
separating out the various functions into physically separate
entities, such as for GUID, SID and LOC.  But that doesn't mean the
only or best definition of the problem is in these terms.

Such a solution necessarily involves every host in fussing around
with changes in the core of the Net.  I see many problems with this:

  Fundamental objections to a host-based scalable routing solution
  http://www.irtf.org/pipermail/rrg/2008-November/000271.html


An alternative viewpoint to your current "root cause" problem
definition is that the current use of IP addresses and the relatively
stable expectations hosts have of them, is just fine.

In this viewpoint the root cause of the routing scaling problem is
the current limitations of the BGP system on its own.  So the core of
the Internet needs a new architectural enhancement so it can provide
multihoming, TE and portability to millions or billions of end-user
networks, while meeting the current expectations of host stacks and
applications that an IP address remains stable, including during
multihoming service restoration.

Such architectural enhancements don't necessarily involve any
locator-ID stuff - for instance some major souping up of BGP in
principle could do the trick - but this is agreed to be impractical.

It so happens that (in my opinion) the best way to do it is with a
map-encap, or map-forward, or whatever, which does do some locator-ID
stuff internally, although invisibly to the hosts.  In this model,
the problem is not in the hosts, or the way they use IP addresses.
It is in the limited capacity of the BGP system to provide
multihomable, portable, prefixes to a billion or so end-user
networks.  LISP, APT, Ivip and TRRP all attempt to provide this
number of prefixes for end-user networks, by means of an additional
architectural layer over BGP.

  - Robin




_______________________________________________
rrg mailing list
[email protected]
https://www.irtf.org/mailman/listinfo/rrg

Reply via email to