On Distributed Communications Series
III. Determination of Path-Lengths in a Distributed Network
V. Edge Binding
In any network, much of the routing power of peripheral stations is wasted simply because peripheral links are unused. Thus, messages tend to reflect off the boundary into the interior or to move parallel to the periphery. Providing alternate paths by edge binding, as illustrated in Fig. 7, should tend to shorten average path-lengths measurably. Figure 8 exhibits flow patterns produced by simulations of a 14x7 network of redundancy-three with and without edge binding. Each diagram is a spatial representation of the network, the entry in position (i,j) of the representation indicating the number of messages routed through the corresponding station since the start of the simulation. The effect of edge binding in reducing interior clustering is clear. Figure 10 exhibits the distributions for the edge-bound net, while Fig. 9 gives statistics for the various runs. Although edge binding reduces clustering and results in an increase in flowrate, it seems clear that the resultant distributions are less desirable than those obtained without edge binding. Apparently edge binding tends to overly penalize paths through the center of the network, and seems to provide higher flow rates at the expense of slightly higher drop-out rates.