Computer Networking Lecture Notes

25 March 2013 • Network Routing

---

Outline

• Networks and graphs.
• Workhorses: distance-vector and link-state routing.
• Hierarchy
• Broadcast and multicast routing.
• Oddballs: anycast, mobile-host, and ad-hoc routing.

Hubs and Bridges

• A hub combines many LANs into one LAN.
• A bridge combines many LANs into a group of LANs.
• A hub LAN could be a network.
• A bridge group probably is a network.
  • Unified host addressing, like a hub LAN.
  • Separate LAN domains, unlike a hub LAN.
• Separate LAN domains give scalability and structure.

Bridge Networks

• Are we done? Do bridges carry us to where we want to be?
• Ultimately, bridge networks are unscalable at large sizes.
  • All hosts addressable at every bridge.
  • Network-wide bridge coordination.
  • Bridge networks leave resources on the table.
• Bridge networks work in small scales (≈ campus), less so in larger scales.

Network Routing

• Scalable networking requires a better kind of bridging.
  • Remove host size dependencies.
  • Exploit available transport resources.
  • Keep structuring and domain separation.
  • Keep coordination within structuring.
• These are hard problems, the network layer starts solving them.
• The network’s answer to bridges is routers.

Flooding

• Be dumb: forward arriving packets on all other links.

  

• The flooding routing algorithm.
• Simple, never fails, no routing tables.
• May run forever, exponential growth.

Dampening

• Drop packets after x hops.
  • x is receiver distance or network diameter.
• Only forward packets once.
  • Tag packets with network-unique ids.
    • (sender id, sequence no)
  • Routers keep forwarded-packet lists.
    • Not close to being scalable.

Comments

• Acyclic network topologies naturally dampen packets.
  • Bus or tree networks.
• Flooding is a common initialization strategy.
  • No routing information required.
• Always finds the quickest route.
  • At enormous cost.
• Failure resistant.

Networks as Graphs

• A network can be represented as a weighted graph.
  • V = { routers }
  • E = { links }
  • W_V = resources, throughput, …
  • W_E = bandwidth, delay, traffic, …
  • These components vary over time.

Routes and Paths

• A route through a network corresponds to a path through a graph.
• A route cost corresponds to a path cost.

  cost(v1, …, vn) =
  		\(\Sigma\)1 ≤ i ≤ n WV(vi)
  	+	\(\Sigma\)1 ≤ i < n WE(vi, vi + 1)

• Costs are often (should be) optimal (minimal or maximal, for example).

Optimality

• Another property of optimal paths:
  • Every subpath of an optimal path is also optimal.

  

• The set of all shortest paths to a given destination form a sink tree.
• Routing algorithms discover and use the sink tree at each router.

Shortest Path Routing

• Given optimality,
  • build a graph representing the network
  • run a shortest-path algorithm on the graph
  • distribute the sink tree to the routers.
• This is a question-begging algorithm.
  • Building the graph, running shortest path, distributing the result,…
• But other routing algorithms should behave like this.

Distance-Vector Routing

• Distribute shortest path to each router.
• Each router maintains a list (a table, a vector) of triples

  (destination, out-link, cost)

• Periodically routers exchange their lists with adjacent routers (neighbors).
• Each router updates its list using neighbor lists.
  • Iterative shortest-path.

Distance-Vector Updating

• Neighbor n sends (d, *, c_d) to r.
• If (d, *, *) is not in r’s list,

  add (d, out-link_n, c_n + c_d)

• If (d, ol_d, c_d) is in r’s list
  • If r.c_n + n.c_d < r.c_d or r.ol_d = r.ol_n, replace

    (d, ol_d, c_d) with (d, ol_n, r.c_n + n.c_d)

  • Otherwise do nothing.

Convergence

• Routers converge when they all agree on shortest paths.
• Distance-vector routing converges
  • rapidly on good news,
  • slowly (hop-by-hop) on bad news.
• This is the count-to-infinity problem.
  • Which renders distance-vector routine less practical.

Network Topology

• Distance-vector routing fails because it is path-independent.
  • Routers update tables solely on cost.
• Make routers path-aware by learning the network graph.
  • Each router runs a shortest-path algorithm.
• Distribute network topology (with costs), not routing information.
• This is link-state routing.

Link-State Routing

• Periodically do
  1. Discover neighbor addresses.
  2. Determine costs to neighbor.
  3. Broadcast neighbor information to routers.
  4. Add received neighbor information to the graph.
  5. Compute shortest paths.
• This is what the Internet currently does.

Link-State Distribution

• A Link-state packet contains
  • a list of (neighbor, cost) pairs
  • a 32-bit sequence number to control flooding
  • a time-to-live value for corruption and crashes.
• Other refinements make the algorithm more robust.
  • Out-bound buffering to suppress redundancy.

In Practice

• Link-state routing has two main variants:
  • Intermediate System-Intermediate System (IS-IS).
    • From DECnet to ISO OSI to IP.
  • Open Shortest-Path First (OSPF)
    • From the IETF.
• They are, essentially, the same.
  • IS-IS has an advantage in multi-protocol environments.

Hierarchy

• Distance-vector and link-state routing don’t scale with size.
  • Larger networks use more resources for all routers.
• Deal with size using hierarchy.
  • The big network is partitioned into smaller regions (or areas, or zones, or …).
  • The regions are linked together by a new parent area (or zones, or segments, or …).
  • Repeat until satisfied.

Hierarchy Examples

• The U.S. phone system (268 million phones in 2003) is a five-layer hierarchy.
  • The lowest level, the central office, is about 1000 phones.
• The Internet is an n-layer domain hierarchy.
  • Structure determined by authorities in each domain.
• The Lightweight Directory Access Protocol (LDAP, rfc 4511) is an n-layer addressing hierarchy.

Hierarchical Routing

• Doesn’t something have to know about everything at some point?
• Hierarchical routing:
  • If level i doesn’t know address a, move it up to level i + 1.
  • The address doesn’t exist if the root level doesn’t know about it.
• So does the root level know everything?

Broadcast Routing

• Broadcast routing delivers a packet to all destinations.
  • Flooding is address-less broadcasting.
• Point-to-point simulated broadcast is simple and inefficient.
• Multidestination routing is a more efficient point-to-point simulation.
  • Each packet contains multiple destination addresses.
  • Routers duplicate packets as needed.

Flooding Revisited

• A flooding packet p arrives at router r on link k.
  • Sender s must think k is the best way to get to r.
  • R concludes k is the best way to get to s.
  • If p arrives on any other link, it’s a duplicate.
• This is reverse path forwarding.

Hierarchical Flooding

• If level i doesn’t know address a, flood level i - 1.
  • Associate the responding level i - 1 grouping with a.
  • Otherwise, forward a to level i + 1.
    • And remember that too.
• Hierarchical routers learn routing table entries while routing.
  • With some inefficiency and cost.

Multicast Routing

• Multicast routing generalizes multidestination routing.
  • Each multicast address represents an address set of receivers.
  • Address management is separated from multidestination routing.
• The routing algorithm moves from sink trees to spanning trees.

Multicast Trees

• A broadcast creates a spanning tree to all receivers.
• Routers not delivering useful multicast packets can be pruned.
• The result is a multicast spanning tree.
• Pruning algorithm can follow unicast routing algorithms
  • Multicast OSPF (MOSPF) - topology based.
  • Distance-Vector Multicast Routing Protocol (DVMRP) - reverse path forwarding.

Anycast Routing

• Here’s a routing problem: send this file to a color PostScript printer.
  • Or maybe the nearest, cheapest color PostScript printer.
• This is anycast routing.
• Anycast routing algorithms redefine costs in unicast routing algorithms.
  • Each feature provided by a node adds 1 to its cost.

Mobile-Host Routing

• Mobility and IP don’t work well together.
  • IP addresses interfaces, which are presumed stable.
  • Remember network + host IP addressing.
• The basic idea is to define a
  • Home location to represent the to-address.
  • A remote proxy as current host.
  • Connect home and proxy.

Ad-Hoc Network Routing

• Now imagine nothing is stationary, routers included.
  • This does not necessarily imply wireless.
• Two possible approaches:
  • Every node becomes a (part-of a) router.
  • Some nodes are chosen to be routers.
• We will consider these in more detail in sensor and peer-to-peer networking.

Summary

• Networks can be represented as graphs, and routes are graph paths.
• Routing algorithms build and maintain graph models of the network.
• The workhorse Internet routing algorithm is the link-state algorithm OSPF.
  • Although other algorithms are used as appropriate.
• Hierarchy is an important technique for scalable routing.

---

This page last modified on 2012 October 16.