Review routing algorithms Choose the appropriate paths Routing

• Review: routing algorithms. – Choose the appropriate paths. – Routing algorithms • Flooding • Shortest path routing (example). – Dijkstra algorithm. – Bellman-Ford algorithm. • Shortest path routing algorithm is perfect if we know the precise network topology. – How to maintain the global network state information? This is also an important part of practical routing algorithms.

• Distance Vector Routing: – used at ARPANET till 1979, still used in WAN(RIP) • Each router maintains a routing table, where each node in the network has an entry in the table. • Each entry contains two parts, the outgoing line for the destination (node) and the distance (time, cost, whatever) to the destination. • Each router knows the distance to its neighbors • In a certain period of time, each router sends the table to each of its neighbors. • When a router receives a table from each of its neighbors, it updates its routing table, determines the new minimum distance and the outgoing link for the destination.

B C A E D A B C D E (0, -) (4, B) (2, C) (5, E) (2, E) (8, C) (0, - ) (5, C) (3, D) (9, A) (5, A) (3, B) (0, - ) (3, B) (8, A) (4, E) (3, C) (2, C) (0, - ) (3, E) (3, A) (3, D) (2, D) (1, D) (0, - ) Routing table for node A Routing table for node D. Let dist(B, A) = 4, dist(B, D) = 1, dist (B, C) = 10, what is B’s routing table in the next cycle?

• The count-to-infinity problem. A B # 1 1 C # # 2 2 2 D # # # 3 3 E # # 4 A B C D 1 2 3 3 4 3 5 4 5 5 6 5 7 6 7 7 8 7 …. . . E 4 4 6 6 8 Good news propagates fast, bad news propagates slow. Some heuristics (split horizon). But the heuristics do not solve the problem.

• link state routing (currently used in the internet, OSPF): » The distance vector algorithm converges too slow. • 1) discover its neighbors • 2) measure the distance to each neighbor • 3) construct a packet telling all that is learned • 4) send the packet to all other nodes -- flooding • 5) each node computes the shortest path to all other nodes -- Dijsktra algorithm.

B 4 A 2 1 4 2 E 5 Link state packets C Node A: (A, seq, age, ((B, 4), (C, 4), (E, 2)) 6 Node B: (B, seq, age, ((A, 4), (D, 2), (C, 1)) D …. Node E: (E, seq, age, ((A, 2), (D, 5))

• If everything goes ok, every node will have the whole information of the network in the next step, algorithm converges faster than the distance vector method. • The count to infinite problem? • Some of current research on the link state algorithm. – Use broadcast tree instead of flooding – Use local update to the fix the routing table instead of running the full shortest path algorithm – When the send the link state packet » periodical or sudden change?