4/30/12

Link-State Routing
• Advertisement step
– Send information about its links to its neighbors (aka link
state advertisement or LSA):

[node, seq#, [(nbhr1, linkcost1), (nbhr2, linkcost2), …]

– Do it periodically (liveness, recover from lost LSAs)

• Integration
– If seq# in incoming LSA > seq# in saved LSA for source node:
update LSA for node with new seq#, neighbor list
6.02 Spring 2012 rebroadcast LSA to neighbors ( flooding)
Lecture #21 – Remove saved LSAs if seq# is too far out-of-date
– Result: Each node discovers current map of the network
• Build routing table
• Link-state routing – Periodically each node runs the same shortest path algorithm
over its map (e.g., Dijkstra’s alg)
– If each node implements computation correctly and each
• Routing around failures node has the same map, then routing tables will be correct

6.02 Spring 2012 Lecture 21, Slide #1 6.02 Spring 2012 Lecture 21, Slide #2

LSA Flooding Integration Step: Dijkstra s Algorithm

LSA: [F, seq, (G, 8), (C, 2)]
(Example)
6 2
A C F
4 Suppose we want to find paths from A to other nodes

6 7 8
(0) (6) (8)
E
A
6 C
2 F
0 2
4 (10)
B 5 D G
6 7 E 8
• Periodically originate LSA 0 2
• LSA travels each link in each direction B D G
– Don’t bother with figuring out which link LSA came from 5
(6) (13) (16)
• Termination: each node rebroadcasts LSA exactly once (11) (12)
– Use sequence number to determine if new, save latest seq (10)
• Multiple opportunities for each node to hear any given LSA
– Time required: number of links to cross network
6.02 Spring 2012 Lecture 21, Slide #3 6.02 Spring 2012 Lecture 21, Slide #4

1
 4/30/12

Dijkstra’s Shortest Path Algorithm Another Example

• Initially Finding shortest paths from A: B
4
D
19
– nodeset = [all nodes] = set of nodes we haven’t processed LSAs:
– spcost = {me:0, all other nodes: ∞} # shortest path cost A: [(B,19), (C, 7)]
A 11 15 13
B: [(A,19), (C,11), (D, 4)]
– routes = {me:--, all other nodes: ?} # routing table C: [(A, 7), (B,11), (D,15), (E, 5)]
D: [(B, 4), (C,15), (E,13)] 7
• while nodeset isn’t empty: E: [(C, 5), (D,13)] C
5
E

– find u, the node in nodeset with smallest spcost

– remove u from nodeset spcost route
Step u Nodeset
– for v in [u’s neighbors]: A B C D E A B C D E
• d= spcost(u) + cost(u,v) # distance to v via u 0 [A,B,C,D,E] 0 ∞ ∞ ∞ ∞ -- ? ? ? ?
• if d < spcost(v): # we found a shorter path! 1 A [B,C,D,E] 0 19 7 ∞ ∞ -- L0 L1 ? ?
– spcost[v] = d 2 C [B,D,E] 0 18 7 22 12 -- L1 L1 L1 L1
– routes[v] = routes[u] (or if u == me, enter link from me to 3 E [B,D] 0 18 7 22 12 -- L1 L1 L1 L1
v)
4 B [D] 0 18 7 22 12 -- L1 L1 L1 L1
5 D [] 0 18 7 22 12 -- L1 L1 L1 L1

6.02 Spring 2012 Lecture 21, Slide #5 6.02 Spring 2012 Lecture 21, Slide #6

Failures Routing Loop in Link-State Protocol

• Problems:Links and switches could fail
– Advertisements could get lost A B
– Routing loop
• A sequence of nodes on forwarding path that has a
cycle (so packets will never reach destination) path
– Dead-end: route does not actually reach destination B to D is via A.
– Loops and dead-ends lead to routes not being valid Link AD fails. D
• Solution A’s LSA to B is lost.
– HELLO protocol to detect neighbor liveness A now uses B to get to D.
– Periodic advertisements from nodes But B continues to use A.
– Periodic integration at nodes Routing loop!
– Leads to eventual convergence to correct state Must wait for eventual arrival
(see Chapter 18) of correct LSAs to fix loop
6.02 Spring 2012 Lecture 21, Slide #7 6.02 Spring 2012 Lecture 21, Slide #8

2
 4/30/12

Fixing “Count to Infinity” with Path Vector

Distance-Vector: Pros, Cons, and Loops
Routing
• + Simple protocol
• In addition to (or instead of) reporting costs, advertise the
• + Works well for small networks path discovered incrementally by the Bellman-Ford update
• - Works only on small networks rule

Suppose link AC fails.

A L1 L3
C L1 • Called “path-vector”
When A discovers failure, it
L2 L2
E:cost=3
E:cost=X5 sends E: cost = INFINITY to B. • Modify Bellman-Ford update with new rule: a node should
L2
Counting to infinity! E B advertises E: cost=2 to A ignore any advertised route that contains itself in the
L1
A sets E: cost=3 in its table advertisement
L2E:cost=X6
E:cost=4
L1
B L1 L3
D L2 Now suppose link BD fails.
B discovers it, then sets
E: cost = INFINITY.
But what if A had advertised Sends info to A, A sets
to B before B advertised to A?
E: cost = INFINITY.
6.02 Spring 2012 Lecture 21, Slide #9 6.02 Spring 2012 Lecture 21, Slide #10

Path Vector Routing Summary

E: l1; cost=2; path=[CE] E: l1; cost=1; path=[E] • The network layer implements the “glue” that
[CE] achieves connectivity
A l1 l3
C l1 To reach E, come this way
– Does addressing, forwarding, and routing
[ACE] path = [E]
l2 l2 [CE]
[ACE] l2 • Forwarding entails a routing table lookup; the
[CE] E E: Self
[BDE] [DE] l1 table is built using routing protocol
l2 [DE]
[BDE] l1 To reach E, come this way • DV protocol: distributes route computation;
l1 l3
B D l2 path = [E] each node advertises its best routes to
[DE]
E: l2; cost=2; path=[DE] E: l2; cost=1; path=[E] neighbors
– Path-vector: include path, not just cost, in
• For each advertisement, run integration step advertisement to avoid “count-to-infinity”
– E.g., pick shortest, cheapest, quickest, etc.
• LS protocol: distributes (floods) neighbor
• Ignore advertisements with own address in path vector information; centralizes route computation
– Avoids routing loops that count to infinity using shortest-path algorithm
6.02 Spring 2012 Lecture 21, Slide #11 6.02 Spring 2012 Lecture 21, Slide #12