The Routing Over Large Clouds Mailing List Archive by date

["Andrew G. Malis" <malis@nexen.com>]

Grenville,

> What's the problem with nodes not decrementing TTLs, given that
> (as far as I can tell) the same end to end effect will be the
> result of using NHRP ?  (i.e. packets will arrive closer to, or
> at, the destination with TTL that doesn't reflect the actual
> number of IP hops according to IP topology.)

The point I was trying to make (perhaps not as clearly as possible) is
that much thought and effort has gone into specifying the
functionality a router is required to have, and decrementing TTL is
only one example of such functionality.  You can glue a cell switch
together with an IP-layer controller so that packets can be switched
at the cell layer, but if the combination doesn't fulfill a router's
functional requirements, then please don't call it one, or try to sell
it as one.

With regard to TTL, CSRs, and NHRP - decrementing TTL guarantees that
if routing screws ups, packets won't loop forever.  If you have a box
that participates in IP routing, but TTL doesn't decrement when
packets traverse that box, then a bug in the routing code can result
in infinitely looping packets.  Thus, a true cell-switching router
(with the emphasis on router) decrements TTL, UNLESS it doesn't
participate in the IP routing algorithm.

NHRP does not prevent TTL being decremented, but since it results in
router hops being skipped, it is just decremented less than it
otherwise would have been.  TTL will still go down to zero if a loop
occurs, since the loop cannot be stictly confined to the cell
switching layer.

Cheers,
Andy