LAN traffic analysis and effective bandwidth calculation.

摘要

In this thesis several months of traffic on the UMass Lowell network is analyzed. The flow patterns are found to be fairly predictable over a daily period. Analysis of the network traffic based on the source and the destination ports demonstrate several features in the access pattern and the applications used, that repeat over each working day. These include the peak traffic hour (12:00 pm:1:00 pm) and the applications generated in this time period. The daily traffic is modeled using a finite state Markov chain and queue performance is analyzed using the concept of effective bandwidth (EBW). This measure represents the link capacity required to maintain the packet loss or delay below a prescribed threshold. The performance of the real-time EBW algorithm for the LAN traffic is investigated. The origin of multiple time scales in traffic, its relation to packet level and burst level queuing and its impact on EBW estimation is examined. It is shown that the EBW provides a good approximation at utilizations when the queue delay is primarily a function of the bursts occurring at the peak rate. At low and moderate utilizations when the queue is a function of both packet level and burst level queuing, the EBW approximation overestimates the probabilities of the queuing delay distribution.