K 10
svn:author
V 5
andre
K 8
svn:date
V 27
2012-11-08T22:40:24.086992Z
K 7
svn:log
V 1498
Save this for later continuation: Different approach on ECMP
(equal cost multi path) routing.  Instead of having shadow
rtentry's linked behind the one in the trie, just have an
array of different egress interface and next-hops.  This
simplifies the code significantly.

A normal add to the routing table then sets the main next-
hop.  A ECMP aware add or modify can change any of the other
next hops in the rtentry.

The next-hop list is sorted by priority (distance in Cisco
parlance) with the highest first.  Equal priority next-hops
are next to each other sorted by next-hop IP address.  If
the highest priority has more than one next-hop load is
equally shared among them.  All other routes with lower
weight are not used.  When a higher priority next-hop is
removed or flagged as unavailable (eg. interface link down)
the next higher priority prefix will become active.  When
only unavailable next-hops are in the rtentry it is ignored
and a less specific match is searched for.  If no route is
found the lookup will fail.  When an unavailable next-hop
becomes available again, the rtentry is valid again and will
match on lookups.  This way we can implement suppression of
routes on link state down interfaces without having to actually
remove the rtentry.  Additionally it is automatically reinstated
when the link comes back.  This way for example a OSPF route
can take precedence and NLRI reachability is ensured.  This
is the same behavior as of Cisco, Juniper and other router
vendors.

END