III. AGGREGATED NETWORK TRAFFIC CHARACTERISTICS ON SERVER
In this section, we evaluate the aggregated network bandwidth usage as on the Microsoft Office Communications Server. Initially, we investigate the overall traffic characteristics for the three periods September, November and December. We illustrate the different traffic for the server view in Figure 1 for the timescales of 1, 10 and 60 seconds. We observe that most of the traffic in bps remains under a 10 Mbps level, with exceptions that occur sporadically for all measurement periods. Once aggregated and smoothed, we observe that in aggregates of 10 seconds, the traffic sparingly reaches 100 Mbit, with most of the traffic being under 10 Mbit per ten seconds. We furthermore note that the traffic for the November measurement exhibits the highest “spikes” of network activities. With additional traffic aggregation to minutes, we observe that the November trace exhibits distinctive and sustained periods of extreme traffic, whereby for multiple minutes (and even up to hours) the overall traffic is more than 60 Mbit/minute.
We illustrate the autocorrelation coefficient characteristics for the aggregated server traffic in Figures 2 for the timescales of 1, 10 and 60 seconds. We observe that the November measurement exhibits a significant level of self-similarity (as measured by the autocorrelation) for the bandwidth per second and ten seconds, and a slightly reduced level for the minute time scales. For the September and December measurements, on the other hand, we note an initially sharp decline in the autocorrelation coefficient, which rises again after about one minute in the bandwidth per second evaluation. Although the December level of autocorrelation is slightly higher than observed for the September one, both are close in their overall behavior. Similar observations can be made for the time scales of ten seconds and one minute, whereby we observe that the autocorrelation exhibits (i) an oscillating decline for the time scale of ten seconds and (ii) a nearly exponential decline indicating the absence or low level of self-similarity for the time scale of one minute.