Tapas: a research paradigm for the modeling, prediction and analysis of non-stationary network behavior
Almudena Konrad, Anthony Douglas Joseph · 2003
This thesis introduces two research methodologies for the efficient development of models and performance analysis techniques to be applied to network measurements. We have developed a methodology for the performance analysis of multi-layer networks. We argue that for the correct evaluation of today's networks, we must draw performance conclusions from a detailed study of the cross-layer protocol interactions. In particular, we have studied the interactions between the Transmission Control Protocol (TCP), a reliable end-to-end transport layer protocol, and the Radio Link Protocol (RLP) [19, 21], a reliable link layer protocol for the wireless connection in the Global System for Mobile communications (GSM) network. Each protocol has its own error recovery mechanisms and by studying the interactions of these protocols, we can improve the performance of the wireless GSM system. We have developed a multi-layer tracing tool to analyze the protocol interactions between the layers. Initially, we hypothesized that delay introduced by RLP would unnecessarily trigger TCP's congestion control algorithm, thus degrading performance. However, our studies have showed this to be false. Using our multi-layer analysis tool, we have identified some of the causes that degrade performance, such as (1) inefficient interaction with TCP/IP header compression, and (2) excessive queuing caused by overbuffered links. The second methodology consists of the preconditioning of data measurements to fit traditional mathematical models. This methodology emerges from the fact that network behavior experience complicated patterns and time-varying path characteristics due to internal network components. Applying traditional modeling techniques to this complex data generates poor models. We have developed two data preconditioning modeling techniques and present a novel approach that enables network researchers to quickly select the most accurate modeling and analysis method for a given wired or wireless network path and network characteristic of interest (e.g., delay, loss, or error process). We show that traditional modeling approaches, such as Discrete Time Markov Chains (DTMC) are limited in their ability to model time-varying characteristics. This problem is exacerbated in the wireless domain, where fading events yield extreme burstiness of delays, losses, and errors on wireless links. We present a wireless simulator (WSim2) that provides the two data preconditioning models developed from our modeling methodology and a feedback algorithm to inform the application about events at different network layers.