Latency and Routing Efficiency Based Metric for Performance Comparison of DHT Overlay Networks

摘要

The general method available for comparison between the Distributed Hash Table (DHT) based structured overlay networks are average latency per successful lookup and percentage of successful lookups. For a DHT based lookup service with lookup retries for failed lookups, both average latency per successful lookup as well as percentage of successful lookups are key to evaluate the performance. Although, average latency per successful lookup may be less for a DHT overlay protocol, failure of lookups at high churn rates may degrade the protocol's performance if its routing efficiency is less. On the other hand, just routing efficiency is not sufficient to measure the performance of a DHT since an overlay network having higher routing efficiency may not have a good performance due to high average latency per successful lookup. We have developed a metric for DHT comparison which incorporates both average latency per successful lookup and routing efficiency in scenarios where applications do lookup retries. We have also modeled three different timeout mechanisms while developing the performance metric. The paper also incorporates a model where the probability of failure of a lookup is increasing at every retry of lookups.