High-speed network traffic: characterization and control
Nina Taft-Plotkin · 1994
This dissertation studies three aspects of high-speed network traffic: traffic characterization, traffic control, and the distribution of traffic over multiple networks links. Our study pertains to ATM and Frame Relay networks, both of which are based on virtual circuits. We begin with a presentation of traffic and virtual circuit models which are used throughout the rest of the thesis. We model an ATM virtual circuit as a tandem queueing system, and study traffic patterns at successive node outputs. We describe the scattering and clustering of ATM cells by determining the entropy of traffic streams. In order to explore entropy as a traffic descriptor we developed and implemented an entropy estimation algorithm which modifies Lempel-Ziv data compression to compute entropy rather than compress data. In addition to entropy, we also study the correlation structure of node outputs and the queue sizes at each successive queue. Next we study a system in which traffic from a single source is distributed over multiple ATM virtual circuits connected in parallel between the source and destination. This type of connection introduces a resequencing problem. The input to the resequencer comes from many network outputs, one output from each ATM virtual circuit, and is not a Poisson process (an assumption made in previous solutions to the resequencing problem). We compute the resequencing delay for this system in two ways: by an exact numerical solution and by an approximation. A traffic source can alter the rate at which it inputs traffic into the network according to feedback information from inside the network. The feedback is received in the form of a few bits attached to traveling cells. The virtual circuit is described using a stochastic fluid model which captures the feedback, and the buffer and bandwidth resources received by a single virtual circuit. The source uses an estimation algorithm to estimate the status of the congested resource along its virtual circuit. A flow control policy based on these estimates is defined. To measure the effectiveness of our policy, we evaluate cell loss probabilities as a function of the feedback delay and the number of feedback bits.