---

Link Failures and Distance Vector Routing

---

• The following example shows how the distance vector adjust the routing tables when there is a link failure...

  You will see some interesting phenomenoms that I will highlight...

• Let's first consider the routing table content after the Distance Vector Routing algorithm has converged:

  

  • The numbers in red circles are the minimum cost (distance) from the node that own the table to a specific node in the network.
  • For example, from A's table, you can see that the minimum cost to G is 2 and you should go to F (from A) to reach G with cost 2.

• Suppose link FG fails and node F detects the failure (when G detects the failure, it will do the same thing - but I will not discuss this case, because you won't learn anything new):

  

  • Node F first set the cost to reach node G via node G to infinite
  • As a result, the route to G changes from:

         To G     Next: G
     

    to:

         To G     Next A
     

  • Node F schedules an update message ([F,G,3] - see figure above. to all its (reachable) neighbors. Thus, the new cost to node G is 3 is conveyed to F's neighbors.
  • F's update message will affect an entry in A's routing table (see figure above).

• Pause a minute and examine the current situation:
  • Nodes accumulate updates for a short period of time before transmitting them to improve efficiency (reduce the number of packets - headers are very large).
  • So the situation above will persist for a certain period. Question is, can it cause harm ? Well....

    

  • Suppose node B was sending data (at a high rate) to G....
  • According to B's routing table, it must send them to A
  • According to A's routing table, it must send them to F
  • Now, according to F's routing table, it must send them back to A
  • According to A's routing table, it must send them back to F
  • And so on... (There is a routing loop in the network.)
  • With the current data speed of Gbps, if the loop exists for longer than 1 second, the amount of packets that will be routed back and forth between A and F will exceed 1 Gbits or 100 Mbytes.
  • It will easily overflow any amount of buffer space available in these routers...

• When node A receives F's update message, it will then update its routing table as follows:

  

• There is a cost change in A's routing table, namely:

     Was:                   Changed to:
  
     To    Cost             To    Cost
     -----------            ----------
      G     2                G     3
  

  Node A must schedule an update message for this route change. Node A will send the new distance to node G "(A, G, 3)" to all its neighbors as follows:

  

  • The update messages are sent to neighbors B, C, E and F.

  • Each neighbor will update its own routing information table.

  • Three of the 4 updates (nodes C, E and F) do not cause any route changes (the routing direction towards node G does not change). In such case, there is no need to schedule update messages.

  • A's update message does cause a route change in node B.

    However, it does NOT cause the cost from B to G to change... because B can still reach G with cost 3 (via C !!!).

    So B will NOT schedule an update message

• Just to make a point: A's update message DOES NOT cause B's cost to G to change !!!

  B's mininum cost to G:
  
     Was:                   After A's update
  
     To    Cost             To    Cost
     -----------            ----------
      G     3                G     3
  

  The following figure shows the futility for B to send an update message (B, G, 3):

  

  • B's neighbor nodes will NOT change the values in the distance tables as a result of B's message (so why send it...) !!!