My major interest in measurements and analysis is focused on gathering information for network engineering. Network measurements are desirable because they allow the network provider to determine the quality of service attainable and plan for growth. Long-term trending analysis of traffic flows, buffer capacity, and link utilization can go a long way toward intelligent network design. Currently these measurement are relatively easy to take and are being used to monitor the NAP's ability to switch the offered load.
However, recent measurements on NAP the have shown two interesting trends. First, traffic has continued to grow at an aggressive rate and will soon approach one terabyte per day. Second, the bursts of the traffic can be very dramatic with variations of 50% seen over five minute samples. These trends could be indicative of the evolution of bandwidth hungry applications such as audio and video transport or may be caused by the shear number of users on the network.
Accurate measurement of the traffic generating these flows would be helpful in determining the best course in providing a high quality of service on an on going basis. They may provide information for link capacity planning, protocol development or protocol segregation. One possibly outcome of this work would be a sort of network seismograph that could measure disturbances and provide postmortem information.
An additional troubling technical issues is that of cell based technologies adding complexity to detailed network measurements. Only byte counts on UNIs and VCCs are currently possible. More detail analysis would require multiple RMON like devices or switch multicast redirecting traffic into a collector.
Legal issues are also very important and a clear understanding of what measurements are allowable is necessary. At the lowest level providers can determine link availability performance metrics. However, more interesting measurements, listed in increasing order of privacy invasion, would be:
These metrics are valuable because they provide incite into baseline cases that can be used to construct next generation network equipment. For example, by counting the number of TCP connections, one can draw a correlation to SVC call setup messages, call hold time, and switch capacity. It information is invaluable should the underling network be replace with a connection oriented fabric.