|Novel Approaches to the Monitoring of Computer Networks|
|Prev||Chapter 5. Tracking Network Growth||Next|
As was mentioned at the beginning of this chapter, the chief aim of this application was to enable Rhodes to determine the growth rate of computers on various parts of its network. This information is important for a number of reasons: It allows the University administration to predict the number of computers that will join the network in a given time period and to budget for resources, such as bandwidth, accordingly. It also provides an indication of which areas of the network are most used, and allows for planning of strategic upgrades of the backbone infrastructure to handle this additional demand.
Information on the number of hosts on various subnets at Rhodes was gathered for just over a year, and the results of this experiment are summarised for a sample of subnets on a monthly basis in Table 5-1.
This table shows the number of hosts on Rhodes' campus as a whole (all), as well as four of the thirty-two subnets on campus — namely the admin subnet (Grounds and Gardens, Estates division and Sports Administration), the hamilton subnet (housing the departments of Computer Science and Information Systems), the resnet subnet (including all student residence networking on campus), and the campus subnet (containing several of the public laboratories on campus). Three of these subnets have already been mentioned in Section 5.4, and will now be examined in more detail.
Table 5-1. Number of hosts on Rhodes University's network
|raw growth rate||26.224||0.280||7.810||6.266||3.630|
A cursory glance at the figures presented in Table 5-1 shows a number of trends. On the campus as a whole, the number of IP addressable hosts has increased over the year. The drop in the number of hosts over June and July is explained by the University's mid-year vacation.
The resnet subnet shows a large growth in numbers from the beginning of the year. This is explained in part by the fact that students vacate their residence room over the end of year vacation, and in part by a significant increase in the number of network-connected residences on campus.
Both the campus subnet and the hamilton subnet show sudden growth spurts over a short period of time. This growth occurs as large laboratories of computers are brought online. In the case of the hamilton subnet, this was two new undergraduate laboratories in January. On the campus subnet a number of new laboratories have been commissioned over the course of the year.
One of the few subnets on campus that has shown very little increase in the number of network addressable hosts is the admin subnet. This subnet also seems immune to the inconsistancies introduced by the University's terms.
Analysis of this data more clearly shows this proclivity. A chart of the number of hosts versus time can be plotted, and the slope of this chart will give an indication of the growth rate. Normal regression techniques can be used to calculate the average slope of this chart, which provides a corresponding growth rate. Accurate regression requires more points than are presented in Table 5-1, so this data was recalculated on a bi-monthly basis.
Unfortunately the presence of University vacations in the data skews the resultant slope. The student population at Rhodes drops significantly over these periods, and this has a corresponding effect on the number of computers in use at the University. The resnet subnet, for example, is almost completely devoid of computers during the December vacation. These drops in numbers and the resultant extreme outliers are caused by a known phenomenon, and so they have been excluded from the chart shown in Figure 5-2
The chart above shows the growth rates for three subnets on campus. The red points represent data collected during the University vacations, and were excluded from the calculation of the slope for each series. The campus subnet is typical of the growth rates seen at the University, whereas the resnet subnet shows a particularly high growth in this area. This relatively steep slope is backed by empirical evidence and is likely to continue as more residences get connected to the University's network.
The hamilton subnet shows one of the inherent traps in analysing this data. Hamilton Building was only finished in late 2001, and only obtained full occupancy at the beginning of the year. This can be seen by the significant differences in the numbers of hosts between week seven and week fifteen. There are two slopes plotted for this series — the solid line represents the slope taking into account all the available data, and the dashed line represents the slope taking into account only data obtained during 2002.
The noticeable difference between these two slopes emphasises the necessity to take into account known events and abnormalities that occur during the course of monitoring. It is only by considering these events carefully that a true picture of the growth rate can be achieved. This sudden increase in numbers as the building gained occupants would become less noticeable as the period of monitoring increased. Unfortunately, the year's worth of data available for this analysis is too short a period to smooth out minor disruptions in the long-term trend.