CS 535, Telecommunications

Quiz 4, 3 November 2004


  1. Explain how it is that circuit switching can be implemented over packet-switched networks.


    Each packet contains a circuit (or channel) identifier. Each routing node contains a table mapping circuit identifiers to outgoing links. Entries are added to the table during circuit set-up as a control packet winds its way through the network; resource negotiations for the circuit may also be going on at each node. Once established, packets flow through the circuit, the mapping table makes next-hop look-up much simpler and quicker than it would be for datagram routing. In the end, a tear-down packet traverses the circuit, removing the circuit's table entry from each node and returning the circuit's resources to the node.

    Most of the answers were concerned with what virtual circuits were, rather with how they were implemented. Packet trains were also a popular answer, but without an indication of how packet trains are formed and maintained.


  2. Which would you expect to provide better support for congestion control: datagrams or virtual circuits? Explain your answer in terms of the outline given in Jacobson's congestion paper.


    Virtual circuits, due to circuit set-up and negotiation, represent a much more predictable packet flow than do datagrams. Predictability is good, because it lets sources confidently send all their packets without slow-start (but within their negotiated limits), obtain stable round-trip-estimates without resorting to more complicated variance calculations, and hand-off congestion control to elsewhere in the network because senders operating within their negotiated limits are blameless.

    On the other hand, the previous paragraph was assuming stable, stress-free operation, which is reasonable (or at least simpler to deal with). Under break-down and over-stressed conditions, the greater flexibility allowed by datagram routing makes it more able to adapt to changing conditions than virtual circuits would be. The counter-argument is that by relying on predictable flows, virtual circuits push the point of collapse further away than it is when datagrams are routed (This counter-argument gets its force from the power of exponential growth: the relatively slow and stable change in routing tables is no match for the exponential spread of congestion. It's similar to those spectacular multi-car collisions that occasionally happen on foggy highways.)

    A majority of the answers cast their lot with datagrams and their flexible routing over virtual circuits. Interestingly, several of the answers that cast their lot with virtual circuits did so with the argument that virtual circuits can use slow-start, variance estimates, and exponential back-off more effectively given the clearer and more meaningful signals a virtual-circuit network can send to the end-points.



This page last modified on 14 November 2004.