On Distributed Communications Series
III. Determination of Path-Lengths in a Distributed Network
VI. Variations in Traffic Density
Techniques for smoothing out fluctuations in traffic density are discussed in ODC-II. A technique suggested and evaluated in that Memorandum uses a modification of the HO-table updating algorithm which allows the values of Hi(j,k) to adapt to changing traffic conditions. The adaptation algorithm is described in the footnote to Fig. 12 of the present Memorandum (p. 37), which considers the effects of lengthy bursts of activity from a single station or between a pair of stations. Such effects can be studied by appending to the simulator provisions for treating specified stations as message sources or sinks, with specified weights associated with each station so chosen. A station specified as a source (sink) of weight k < 1 will be chosen by the simulator as a message originator (addressee) with probability k. Since correct receipt of each message block is acknowledged by the adjacent receiving station, it is clear that any prolonged increase in the rate of message generation by a single station will result in a commensurate increase in network loading. In such cases we would expect effects similar to those produced by an increase in message loading. In cases where many stations suddenly increase their rate of message generation to a particular station, a similar increase in network loading should occur. This is prevented from occurring by mechanisms described in ODC-VI II.
Several hog cases were simulated on a 7x7 network of redundancy three. The procedure used was the following: first, simulate the network at a 20 per cent loading ratio without sources or sinks; then increase the loading ratio to 40 per cent, specify a commensurate source/sink configuration, and continue the simulation, comparing statistics at each step. The results are summarized:
- Regions of heavy flow activity centered about sources and sinks;
- Source/sink configurations caused a 50 per cent decrease in the message-flow rate- -this was probably caused by input-choking;
- Distributions of stack waiting times--with sources and sinks--were identical to those obtained for 40 per cent loadings without sources or sinks;
- Single stations acting as both source and sink had little effect on the distribution of path-lengths of delivered messages, the distributions being only slightly less desirable than those obtained for 20 per cent loading with no sources or sinks;:
- Two stations, each acting as both source and sink, had effects on the path-length distribution which depended on the relative positions of the stations within the network--the more remote the stations, the worse the distributions.
We conjecture that single stations acting as a hog source will reduce message-flow rate but will result in few, if any, dropped messages.
The case of a single source was simulated, such as would occur when a "fraudulent" station attempted to overload the network. The results were as anticipated--the increased loading produced distributions expected from the new loadings. Single sources had essentially the same effects as single source/sinks, with the exception that stack waiting times remained relatively unchanged.