EIGRP Convergence

EIGRP Convergence
Let’s take a look at the convergence time of Enhanced IGRP (EIGRP). We will again use Figure 1.2
to help describe the convergence process:
1. Router D detects the link failure between Routers D and F and immediately checks its
topology table for a feasible successor. We will assume Router D does not find an alternate
route in the topology table and puts the route into active convergence state. In reality,
taking bandwidth and delay into account, all that must be true for the path through
Router E to be a feasible successor and for the process to stop right here is for the metric
from Router E to the Ethernet segment off of Router F (the reported distance, RD, from
Router E for this route) to be less than Router D’s metric of the route that just went down
(the feasible distance, FD). If this is the case, this path will be in the topology table, and convergence
will already be over. I give you the beauty of EIGRP. Beware the ugliness, of which
being a proprietary protocol is a good example.
2. Router D sends a QUERY message advertising the routes that it lost with infinite metrics
(4,294,967,295, same as for IGRP) out all active interfaces looking for a route to the failed
link and affected networks. Routers C and E acknowledge the QUERY.
3. Router C sends back a REPLY message advertising the routes requested with infinite metrics.
Router D acknowledges the REPLY.
4. Router E sends back a REPLY message with routes to the networks that Router D lost, including
to the downed network, thinking that it still has the alternate, equal-cost route to offer, not
yet having been informed by Router F of its demise. Router D acknowledges the REPLY.
5. Router D places the new routes in the topology table, which then updates the routing table,
due to their unopposed selection as successors.
6. Because both neighbors sent back REPLY messages, Router D sends both of them
UPDATE messages, thinking that its local routing table has settled down with the
changes. Because Router C has been using poison reverse, Router D updates it with the
two new routes it learned. But because it learned these from Router E and has nothing
new from Router C, the UPDATE to Router E is blank. Return UPDATE messages are
often considered acknowledgments for earlier UPDATE messages, with no separate
acknowledgment messages necessary.
7. Router C responds with an UPDATE, which Router D acknowledges. Router E sends an
UPDATE with the link between Router D and Router F, which it still thinks is accessible
through Router F. The reason Router E includes it in an UPDATE message is because it
once thought there were two equal-cost paths to get there. Any such changes are eventually
sent in an UPDATE message, because these are the messages that suggest the dust has settled
and these are the results.
8. However, shortly thereafter, Router E learns from Router F that the network between Router D
and Router F is truly down, and immediately sends out a QUERY to Router D looking for
another path to the downed network, which also serves to notify Router D that the network is
inaccessible through Router E now, as well. Router D acknowledges the QUERY.
28 Chapter 1  Routing Principles
9. Router D sends a REPLY, which is acknowledged by Router E, advertising that it too has an
infinite metric to that network. Router D and Router E now consider themselves synchronized.
Router D also updates its own EIGRP tables with the fact that the network is lost, but
it still knows that the Ethernet segment off of Router F is accessible through Router E.
10. In response to this latest news from Router E, Router D sends a QUERY to Router C, just
to make sure that it hasn’t learned of this network in the meantime. Router C acknowledges
this QUERY and sends back a REPLY message that confirms to Router D that no path
exists to the downed network. After acknowledging this REPLY, Router D and Router C
consider themselves synchronized. The EIGRP routing domain has converged.
Router A convergence time is the total time of detection, plus the query and reply time,
plus the update propagation time—about two seconds total. However, the time can be
slightly longer.
In case it was not apparent from the foregoing discussion, EIGRP employs
various message types, including UPDATE, QUERY, and REPLY. It makes
sense that UPDATE and REPLY messages can carry new routing information
that could alter the receiving router’s EIGRP tables. More subtly, QUERY messages
also carry route information that the receiving router treats as new,
with respect to the sending router. The QUERY/REPLY pair is invaluable to
EIGRP to make sure that one or more specific routes are synchronized
between the pair of routers exchanging these messages, especially when
there was an earlier discrepancy. For example, in step 8 in the preceding
description, Router E used a QUERY message, not an UPDATE message, to
inform Router D that it agreed that the link between Router D and Router F
was down.