ROUTING
-------------

Definition:
The selection of paths through which a network sends packets.  For an overview, see  http://www.freesoft.org/CIE/Topics/87.htm

The network must know the location of a destination before selecting a path which leads to it. Networks learn that certain paths are conjested or disabled and uses that information to improve the path selection.

A user normally specifies a destination by name (ogimaa.nmu.edu), rather than by giving the location of that destination on the network. Since the name doesn't tell you exactly where the computer is, the network translates the name into the location.  This translation can be done by computing the network portion of the IP address.

Routing is physically taken care of in the Network Layer of the OSI model.

Who routes?

• Source  Puts the work on the source and not the routers (bad for your Commodore 64).  Requires every source to be informed of topology changes.  Only works with virtual circuits.  On the internet, happens only if the packet has either loose source route or strict source route  option set.
• Routers Requires more work by routers.  Can use either virtual circuits or datagrams.  How the internet works.

• Session Routing

Path through which all packets of the same conversation are sent. Therefore, there's one routing decision per conversation. This is less computation, but requires the routers to maintain state and does not react to changes in the network during a conversation.  This is sometimes called session routing, static routing, or non-adaptive routing.
• Datagram Routing

Makes routing decisions for each packet.
More work for each router, but no need to maintain any state, and reacts quickly to changes.
Sometimes called adaptive routing.

• Correctness The algorithm NEVER screws up, even for years of service and even during hardware, software, and human failures.
• Robustness The algorithm must adjust the routing decisions when equipment fails and when traffic conditions change. A robust algorithm monitors the network constantly and updates the routing decisions either frequently or soon after a change is detected. Basically, it monitors the network, makes routing decisions, and implements these decisions. Routing decisions should be modified before congestion levels increase significantly.
• Stablity A routing algorithm is stable if the routing decisions adapt smoothly to changes in operating conditions. For example, a small change in operating conditions should result in a comparatively small change in routing decisions. Instability could result from the effect of routing decisions on the  operating conditions.
• Fairness A routing algorithm is fair if it results in similar delays for the packets of different sources and destinations. For example, routing is fair if the average delay per packet is the same for two separate streams of packets being transmitted.
• Optimality A routing algorithm is optimal if it maximizes the network designer's objecting function while satisfying design contraints.   But what should you maximize.  Throughput?  Latency?  Number of hops traversed?

Note: Optimality and Fairness aren't always compatible. Such incompatibility may be caused by an objective function that doesn't properly reflect the
operating cost of the network.

Examples of widely used routing algorithms:

• Flooding Every node in the network transmits a copy of each packet it receives on every one of its transmission links, except on the link through which it received the packet.  Basically, it tries EVERY route for EVERY packet.  Very costly, but if a route exists it will be found.
• Spanning Tree Routing The transmission of packets along a tree of paths that leads to a specific destination.  If everyone uses Shortest Path, you will get a Spanning Tree.
• Routing Table Each node transmits a packet along a link selected according to some probabilities that depend on the destination.  The sysadmin writes the table, using his best judgement.  Works only for small networks.
• Bellman-Ford, or Distance Vector Determines the fastest path between nodes from the travel times along individual links.  But it suffers from Counting to infinity, which HAS NO POSSIBLE SOLLUTION.  In practice there are many ad-hoc ways to solve this, including split horizon. See  http://www.freesoft.org/CIE/RFC/1058/6.htm
• Hierarchical Routing Sizes of the routing tables are significantly reduced by using this algorithm. The nodes are grouped into domains, and each node keeps a list of the next node for each other domain and for each other node in its own domain.
• Shortest Path Routing  Route every packet along the shortest path, where path length is defined by (latency, # of hops, cost, etc.) Use Dijkstra's algorithm! Much more can be found at http://renaud.waldura.com/doc/java/dijkstra/. An animation can be found at animated demo.