distance vector algorithms

[Martha Steenstrup <[e-mail]>]

Hi Tony,

The context of the original looping discussion was the IDPR
architecture document, in which I compared the applicability of both
distance vector and link state algorithms to policy-based routing.
Hence, I was never addressing arbitrary distance vector algorithms,
but rather specifically distance vector algorithms for policy-based
routing.

As I mentioned in the architecture document, any such distance vector
algorithm must carry the complete route constructed thus far, so that
intermediate gateways can make decisions whether to accept or reject
the route according to their source and transit policies (the access
restriction portions, in particular).  Hence, in Yakov's terminology,
distance vector algorithms for policy-based routing must be "path
vector" algorithms.  BGP and IDRP are two such examples.  However, my
discussion was not restricted to these two protocols only, but rather
applied to any path vector algorithm for policy routing.

Using path vector algorithms for policy-based routing and applying
route selection criteria based on source and transit policies means
that gateways executing these algorithms do not necessarily accept the
most recent routes that they learn about.  They judge each route
according to its consistency with their policies.  That is exactly as
it should be.  However, such route selection also means the possibility
of the type of looping I described in the previous messages.  Hence,
path vector algorithms for policy-based routing should have increased
loop detection capabilities, to reduce the probabilities of such loops.
Even so, I do not believe that these algorithms can ever eliminate this
type of looping problem.

m