Routing Protocols
==================

* What is routing? Determining the best path to reach a
  destination. Routing protocols form the control plane of the network
  layer, and responsible for coming up with a set of routes for any
  destination, and for picking the best route.

* Any routing protocol roughly consists of two phases: route
  advertisements using which information about destinations is
  exchanged via messages, and best route computation where a best
  route is chosen to be installed into the forwarding table.

* What are the criteria to pick a path? Usually, intra-domain routing
  protocols pick the shortest cost path. Every link has a metric, and
  the path that minimizes the sum of all link metrics is chosen as
  best path. How are metrics chosen? Can be simple hop count, based on
  bandwidth, latency, physical distance etc. That is, based on
  desirability of using the link. Inter-domain routing protocols may
  use other criteria like policy. Network operators may also wish to
  balance load across links (traffic engineering). But usually, the
  simplest way to pick path is shortest path or least cost path.

* Load-sensitive vs load-insensitive routing: most routing protocols
  do not pick best path based on congestion, as it can lead to
  oscillations and instability. That is, link metrics do not depend on
  current load level. For example, if some link is loaded and its
  metric reduces, all flows may move away from it, leading to load on
  another link, and all flows will move back again.

* Static vs dynamic routing: for small networks and/or networks that
  rarely change, manually setting the route once is sufficient. This
  is called static routing. For more complicated networks, dynamic
  routing protocols are used, that adapt routes in response to changes
  in network topology. We will study two classes of dynamic routing
  protocols: link state (LS) routing protocols, and distance vector
  (DV) routing protocols.

* Link state routing: "tell about your neighbors to everyone". Each
  node collects information about all its neighbors and the link
  metrics. This LSA (link state announcement) of every node is flooded
  through the entire network. So, at the end of it, each node has
  complete view of the network graph. Each node then independently
  runs Dijkstra's shortest path algorithm to get the shortest path to
  every destination, based on which it figures out the next hop for
  every destination.

* Suppose a node u has obtained LSA of all nodes in the network. This
  is the algorithm that u runs to find shortest path to all
  destinations.

Let N = set of all nodes
Let D(x) = distance to x from source node u
Let c(x,y) denote the cost of the edge from x to y
Let prev(v) denote the previous node on the shortest path from source node u to v

Initially, let N' = {u}
For all nodes v that are neighbors of u, D(v) = c(u,v)
For all other nodes D(v) = infinity

Now, we will run the following loop. Each iteration of the loop adds
one node to N', and the loop terminates when N' = N.

Loop:
  Find w not in N' that has the least D(w) and add to N'
  For each neighbor v of w that is not in N':
    if(D(w) + c(w,v) < D(v))
       D(v) = D(w) + c(w,v)
       prev(v) = w

* What is the above update rule doing? If u has a shorter path to v
  via w than what it had before, then the path via w is used by u as
  the shortest path to v. Node v stores its previous hop as w, and the
  complete shortest path from u to any node can be found by tracing
  the previous pointers.

* Distance vector routing: "tell about everyone to your
  neighbors". Every node exchanges a distance vector (a vector
  containing its estimate of distance to each destination) with its
  neighbors. Upon receiving a neighbor's distance vector, a node
  updates its distance vector by adding its link cost to neighbor's
  distance vector, and if a better path is found through neighbor, it
  updates its best route. This is called the Bellman-Ford update
  algorithm.

* In DV, every node maintains a distance vector D, which lists its
  estimate of the distance to all destinations. Initially, at a source
  node x, D(y) = c(x,y) for directly connected neighbors, and D(y) =
  infinity for all other y.

Node x now obtains a distance vector Dv from every neighbor v, and
updates its distance vector as follows.

D(y) = min over all v { c(x,v) + Dv(y) }

* Distance vector has "count to infinity" problem. Suppose node A has
  a path to a destination and B is using the path via A. Now if A's
  path fails, A might mistakenly use B's distance vector and start
  using B's path (which goes via A) leading to routing loops and
  invalid paths. Simple solution is "split horizon" or "poisoned
  reverse": if B is using the path via A, then B will announce a
  distance vector of infinity to A for that particular destination.

* LS vs DV: LS has higher messaging complexity but converges faster to
  the correct path.

* OSPF (Open Shortest Paths First) is a popular intra-domain routing
  protocols that uses simple LS routing logic. RIP (Routing
  Information Protocol) is a simple DV protocol.

* Work out examples of LS and DV routing protocols in action.