CS558 Sylabus

The Reverse Path Broadcasting Technique

Recall in the Distance Vector Routing Algorithm, each node sends its "distance vector" to all other nodes to its neighbor nodes in the network ( click here)
The lack of a broadcast mechanism in the distance vector method means that not all the nodes in the network will know which routers serve multicast receivers
So the trick we pulled in the link state protocol (i.e., including multicast membership information in the broadcast link state route messages) does not work in distance vector routing... unfortunately.
(If you wonder why we can't just add a broadcast method to the distance vector method... if you do that, you changed the nature of the distance vector method: you only send information to your neighbors. If you create the capability to broadcast, well then, why not broadcast the link state to every node and use Dijkstra to compute the route ?)
Extending the (unicast) Distance Vector Routing Algorithm for multicast routing is not a trivial matter (as it is the case in the link state routing method) because routers do not have full knowledge of the topology of the network.
But strangely enough, the Distance Vector Routing Algorithm was the first routing algorithm that has been adapted for multicast routing.
In 1993, Steve Deering extended the SunOS 4 kernel (and later other open source UNIX kernel) with multicast multicast capability.
The Sun OS 4 kernel used the Route Information Protocol whichwas based on the Distance Vector Algorithm.

The Reverse Path Broadcasting technique

The place to start learning about how to extend the distance vector protocol to multicast routing is a research paper by Dalal and Metcalfe in 1978 published in the prestigious ACM Communications.
You can get a copy of that paper here: click here
BTW, if "Metcalfe" sounds familiar, you may recall me mentioning that he is the inventor of Ethernet....
The Dalal & Metcalfe paper describe a method to perform more efficient (than flooding - see click here) distribution of broadcast messages
Dalal & Metcalfe saw that the routing tables contains information of all reverse trees
So what the heck is a reverse tree ???
The FORWARD Tree
- The forward tree of a node X consists of all shortest path from X to all other nodes.
- The following figure shows a network and 5 smaller figures that shows the shortest paths from A. B, C, D and E to all other nodes in the network:
The REVERSE Tree
- The reverse tree of a node X consists of all shortest path from all other nodes to X.
- The following figure shows the same network and 5 smaller figures that shows the shortest paths from all other nodes to A. B, C, D and E:
Properties of FORWARD and REVERSE Trees
- When link costs are ASYMMETRIC that the forward tree of node X is (generally speaking) not equal to the reverse tree !!!
- When link costs are SYMMETRIC, the forward trees and the reverse trees are equal
  For example:
  - Forward tree with SYMMETRIC Costs:
  - Reverse tree with SYMMETRIC Costs:
Facts about multicasting:
- In order to do multicast (or broadcast) from a source, we need a tree that is rooted at the source
- It could be any tree, and the reverse tree can serve the purpose
- However... multicast packets transmitted using links of the reverse tree will not follow minimum delay routes from the multicast source to the destinations
- (Because the minimum delay routes forms the forward tree and in general, the reverse tree is not equal to the forward tree).
- But reverse trees is very close to the forward tree....
Reverse Tree BROADCASTING
- Dalal and Metcalfe discovered that the unicast routing information encodes the reverse tree....
  For example, consider the node A in following network and the contend of the unicast routing table that forwards packets towards node A:
  
  You can clearly see that the unicast routing information encodes the reverse tree
  Note that the unicast routing information encode a tree that spans all nodes of the network.
- In the 1978 paper, Dalal and Metcalfe developed an efficient broadcast method that makes use of the reverse path tree
  This method is known as Reverse Path Broadcasting (RPB).
  Since the reverse tree spans all nodes in the network, their method will broadcast... At the end of their 1978 paper, they mentioned that their method could be extended for multicast. And indeed, in 1993, Steve Deering extended their method to multicast.

Reverse Path BROADCASTING: the principle

The basic underlying technique to implement reverse path broadcasting that uses no additional information other than unicast routing tables is:

When a node X receives a (broadcast) message from A, it transmits the (broadcast) messages on all its other links (except on the incoming link) if and only if the (broadcast) message was received on the link that X uses to forward unicast messages to A.

Example Reverse Path Broadcast
- Consider the network in the above figure....
- Suppose node A sends a broadcast message with header information: (To:ALL, From:A)
- This message is forwarded by the basic Reverse Path Forwarding principle as follows:
  1. Node A sends the broadcast message:
    - Notice that node A has no information available about the reverse tree rooted at A !!!
    - Therefore, it must send the message to all its neighbors to ensure that
    - Note that the messages is FROM A and has the source ID = A, this is important for the next steps.
  2. Nodes B, C and D receive A's message. The following figure depicts what each nodes will do with the received message (explanation is given below the figure):
    - When B receives the message, it can detect that the source ID = A
    - Then B checks if the messages was received on the link it uses to transmit to A. It did. So B forwards the message onto its other links
    - C receives the message from A, it can detect that the source ID = A
    - Then C checks if the messages was received on the link it uses to transmit to A. It did not (C uses E to forward to A). So C drops message from A.
    - When D receives the message, it can detect that the source ID = A
    - Then D checks if the messages was received on the link it uses to transmit to A. It did. So D forwards the message onto its other links
  3. Nodes E receives B's and D's messages. The following figure depicts what E will do with the received messages (explanation is given below the figure):
    - E receives the message from B, it can detect that the source ID = A
    - Then E checks if the messages was received on the link it uses to transmit to A. It did not (E uses D to forward to A). So B drops message from B.
    - E receives the message from D, it can detect that the source ID = A
    - Then E checks if the messages was received on the link it uses to transmit to A. It did. So E forwards D's message onto its other links
  4. Nodes C receives E's messages. The following figure depicts what C will do with the received messages (explanation is given below the figure):
    - C receives the message from E, it can detect that the source ID = A
    - Then C checks if the messages was received on the link it uses to transmit to A. It did. So C forwards E's message onto its other links
  Bottom line: every node receives the messages from A.... albeit not as efficient (yet) as it should... that will change with some improvement....
Problem with the basic approach

Although broadcasting has been realised.... it is not realised with the minimum number of messages.

It would not be high quality research worth publishing in the prestigious ACM Communication if Dalal & Metcalfe did not remove these redundant messages....
In order to achive broadcast using the minimum number of messages, the distance vector route computation algorithm must be extended
But remember: any extension must preserve the essence of the distance vector route computation algorithm.
Spelled out in English it means: a node must only send information to only its neighbor nodes. (And don't go extend the distance vector algorithm with broadcast techniques as mentioned above....)

Reverse Path BROADCASTING: the efficient solution

Add new Broadcast Routing Table data structure to every multicast-able routers:

Broadcast Routing table: Broadcast Source | Next routers ------------------+---------------- A | .... B | .... .... | ....

This table will be used to record the branches of the Reverse Tree
The information needed to construct the branches of the reverse tree in not available in the Distance Vector routing method and therefore, this information must be added to the messages that are exchanged by the routers...
What information must be added ??? (Answer is below....)

When a router send distance vector update messages (these are the messages that routers exchange among themselves to when routing cost changes) to their neighbors, the messages include the next node.

Pictorially speaking:

Ordinary distace vector: Dalal & Metcalfe's distance vector: ------------------------ ------------------------------------ (From, To, Distance) (From, To, Next, Distance) Example: Node A ======= Distance vector used Distance vector used by Dalal in unicast routing and Metcalfe: method: From To Distance From To Next Distance -------------------- --------------------------------- A A 0 A A - 0 A B 6 A B B 6 A C 3 A C C 3 A D 3 A D D 3 A E 6 A E D 6

Finally, a procedure to process the additional information:
A node that receives distance vector messages, in addition to the orignal processing (see click here), must also:
- Check if my own ID == Next in the distance vector message.
- If EQUAL, then the router enters the From value into Next Router field of the Broadcast Routing Table for the entry that corresponds to Dest

Example:
For simplicity, I will only show the portion of the distance vector message that contain information affecting the construction of the broadcast tree rooted at A.
I will show the "final" distance message exchanged by the nodes when the route construction has converged. (These message are easily derived from the final routing tables !)
The final distance message from node B that affect the broadcast tree rooted at A is as follows:

The final distance message from node C that affect the broadcast tree rooted at A is as follows:

The final distance message from node D that affect the broadcast tree rooted at A is as follows:

The final distance message from node E that affect the broadcast tree rooted at A is as follows:

The broadcast routing tables clearly shows that broadcast messages from A will follow the reverse tree and it uses the minimum number of messages:
Note: reverse path broadcast tree from other sources are constructed in the similar manner. Try construct one or more as exercise.