Notes On Chapter Eighteen -- WAN Technologies and Dynamic Routing

• 18.1 Introduction
  
  
  • WAN structure
  • Basic WAN components
  • The concept of Routing
  • Routing algorithms
  
  
• 18.2 Large Spans and Wide Area Networks
  
  
  • True WAN technology is scalable - able to grow and deliver reasonable performance, connecting many widely-separated sites, each containing a large number of hosts.
  
  
• 18.3 Traditional WAN Architecture
  
  
  • Refer to figures 18.1 and 18.2 on page 307
  • WANs predate LANs and the Internet
  • Early WANs were designed to connect sites that each had just a few large computers.
  • Early WANs consisted of digital circuits between sites, computers at the sites, and packet switches.
  • Packet switches had interfaces of one kind for accepting inter-site digital connections.
  • Packet switches had other interfaces for accepting slower data connections from computers at the local site.
  • Usually nowadays the functionality of a WAN packet switch is achieved by combining one or more LAN switches with a router.
  
  
• 18.4 Forming a WAN
  
  
  • Refer to figure 18.3 on page 308
  • WANs are designed, taking into account required capacity, data rates, distances spanned, redundancy required, and amount of delay that can be tolerated
  • Long-haul connections between sites may be implemented with various technologies, including wire or fiber leased lines, microwave and satellite channels.
  
  
• 18.5 Store and Forward Paradigm
  
  
  • Packet switches use a store and forward paradigm.
  • Packet switches have memory, and the ability to temporarily store packets they have received in their memory.
  • When a switch receives a packet, it is not required to forward it to another interface immediately. If the interface is busy, the switch can temporarily store the packet in a queue in memory.
  • Because of the ability of packet switches to queue (to buffer) packets, senders do not have to wait for each packet to be forwarded by the switch before sending another packet to the switch. Therefore, overall utilization and performance of the network is supported.
  
  
• 18.6 Addressing in a WAN
  
  
  • Refer to figure 18.4 on page 310
  • In a WAN, typically computers have hierarchical addresses - two-part addresses. One part designates the packet switch to which the computer is attached and the other part designates which one of the computers directly attached to the switch.
  
  
• 18.7 Next-Hop Forwarding
  
  
  • Refer to Algorithm 18.1 on page 311
  • Refer to figure 18.5 on page 312
  • Packet switches use forwarding tables or the equivalent, to determine how to route packets.
  • When a packet arrives at a switch, the switch extracts the first part of the destination address (basically the destination switch number). If the packet needs to be forwarded, the switch does a table lookup to determine the next hop interface - the outgoing connection on which to send the packet so that it can be sent to the next switch along the path it is supposed to take through the network.
  • Note that each packet switch is concerned only with how to determine a packet's next hop. Packet switches don't need to keep track of entire pathways that packets follow.
  • If switch number X receives a packet with a destination address of the form (X,Y), the switch looks up Y in a table or uses some other means to determine on which local interface to transmit the packet. This is how the packet is routed locally to the NIC that is its final destination.
  
  
• 18.8 Source Independence
  
  
  • Next-hop forwarding is source independent
  • A packet switch bases its forwarding decision solely on the destination address.
  
  
• 18.9 Dynamic Routing Updates in a WAN
  
  
  • Refer to figure 18.6 on page 314
  • Refer to figure 18.7 on page 314
  • Because links between packet switches can go up and down, typically routing software continually monitors the network graph and updates the forwarding tables.
  • The job of routing software is to insure that collectively the packet switches route packets along the shortest paths currently available between source and destination.
  
  
• 18.10 Default Routes
  
  
  • Refer to figure 18.8 on page 315
  • If many entries of a forwarding table contain the same next hop designation, say H, one may remove all such entries from the table and represent them instead as a default route to H.
  • To route a packet, first check the remaining part of the table. If the destination switch number does not match any of those table entries, then conclude that next hop H is to be used.
  • This scheme may cut way down on the amount of memory needed to store forwarding table.
  
  
• 18.11 Forwarding Table Computation
  
  
  • Static Routing: a program computes forwarding table when the packet switch boots and it does not change.
  • Dynamic Routing: like static routing except the program alters the forwarding tables as network conditions change.
  
  
• 18.12 Distributed Route Computation
  
  
  • Refer to algorithm 18.2 on page 317
  • Refer to algorithm 18.3 on page 319
  • 18.12.1 Link-State Routing (LSR)
    • Link-State Routing (LSR) - aka Link-Status Routing - aka Shortest Path First (SPF) routing
    • Packet switches periodically broadcast (to all other reachable packet switches) information they know about which links are up or down (e.g. "A link between switches 5 and 9 is up" or "The link between switches 3 and 12 is down").
    • Using this information software running on each packet switch maintains and updates a representation of a graph showing all links known to be up, and the packet switches connected to them.
    • Each time it updates the graph, the software next uses Edsger Dijkstra's "all nodes, shortest paths" algorithm to (re-) compute the shortest possible paths between itself and all the other reachable packet switches.
    • The algorithm assigns a 'weight' to each link - some number that indicates the relative amount of delay or other cost of traversing the link. "Shortest path" means "path with least total weight.
    • After it computes a new shortest-path, the software on the packet switch updates, if necessary, the next-hop entry of its forwarding table corresponding to the endpoint of the new path.
    
    
  • 18.12.2 Distance Vector Routing
    • Periodically a packet switch sends a (DVR) message to all the other packet switches to which it is directly connected.
    • The contents of a message is a list of items of this form:
      
      "I can reach destination X, and its current distance from me is Y"
      
      The list sent by a packet switch S contains such an item for every other packet switch that S can reach.
    • This information is not broadcast. None of the adjacent switches forwards these messages. They just pass between packet switches and their directly-connected neighbors.
    • When a packet switch M receives a DVR message from a neighbor N, it can use the information in various ways to update its forwarding table. (See algorithm 18.3 on page 319.)
  
  
• 18.13 Shortest Path Computation in a Graph
  
  
  • Refer to figure 18.9 on page 320
  • Dijkstra's algorithm may be used with different definitions of weight.
  • Sometimes each connection between packet switches is assigned a weight of 1.
  • Sometimes high weights are assigned by network managers to discourage use of the corresponding links.
  
  
• 18.14 Routing Problems
  
  
  • Routing problems can occur if Link State Routing messages are lost or if a packet switch receives DVR messages that it sent.
  • Problems with DVR can be more severe - e.g. routing loops.
  • Routing systems in actual operation have some safeguards but problems do sometimes occur.