Loss behavior analysis and its application in design of link quality metrics

Abstract

Wireless losses play a significant role in determining the performance of many higher-layer protocols. Various measurement studies have presented results that explore the many facets of link behaviour. However, the results have often been contradictory. In the first half of the paper, we delve into understanding the underlying causes of losses and resolve two contradictions that appear in literature. One of the key insight we gain from this study is how packet loss manifests as a function of the packet count over which the averaging is done. This observation has wide applicability in many research domains that routinely employ loss rate measurements to dictate protocol behavior. To illustrate this point, we focus on a specific research domain: link quality metrics as employed in routing protocols. In the second half of the paper, we show that the popular link quality metric ETX and variants thereof employ incorrect averaging which results in unstable and degraded performance. We modify the averaging mechanism based on our insight and term this modified version SLIQ. SLIQ stands for stability based link quality metric. The modification is simple yet subtle, and the performance improvement it provides is substantial. We carry out our evaluation of these metrics using realistic traces obtained from a wireless 802.11a testbed. We consider the impact the modified link metric has at both the routing layer and the application layer by defining appropriate performance metrics. When compared with ETX and another previously proposed metric ROMA, SLIQ provides stable and persistent routes (cuts the number of route flaps by a factor of 6) and can support twice as many high quality voice calls as ETX.