Latency and Contention

In the 40’s Bob Hope said:

A study of the post office costs reveals that from a three cent stamp, one cent goes towards moving the mail and two cents goes to storing it.

Then as now latency is a principle problem with communication networks. There is a tendency to suspect that the “store” in “store and forward” style networks can be avoided somehow. When one imagines the scenario for an individual packet it seems plausibly possible to arrange that the packet need not wait at the switching points. Perhaps queuing algorithms are merely too simple.

The problem is deeper than it first seems. First we note many analogs to the problem, perhaps better to see its scope.

The Post Office example is classic.
Air Line layovers for traveling between Sacramento and Albany
Sonet ‘time slot interchange’ imposes a small delay associated with sharing a trunk.
Traffic lights in a mesh of city streets impose delay.
and, of course, packet queueing in a digital packet network.
The delay introduced by the filters that extract information from a frequency multiplexed signal.

There are many more.

The nature of the delay is much the same across these examples and the underlying reasons for these delays tend to be isomorphic. Even with weeks foresight the airline layover cannot be avoided. The air traveler puts up with the delay and in turn is assured that an aircraft will be available, even in case of lack of profitable load.

The Post Office and airlines schedule their bulk movements with layover minimization as only one goal. Sonet schedules time slots locally at the beginnings of calls. I think that global scheduling would bring little or no benefit. Switching schedules in mid call introduces jitter which could only be eliminated with a jitter buffer which would nullify the advantages of a better schedule. Sonet delay was seldom a problem.