Two analytical models for evaluating performance of Gigabit Ethernet hosts with finite buffer

Abstract

Heavy network traffic as that of Gigabit Ethernet can negatively impact system performance of hosts. Gigabit Ethernet hosts may livelock as the CPU processing power gets mostly consumed handling interrupts of incoming packets. Two analytical models are developed to capture host behavior and evaluate its system performance. In this paper, we consider realistic hosts with limited-size or finite buffer. The first analytical model makes use of the known finite-buffer Markovian queue ( M/M/1/B), while the second, which is more accurate but more complex, is a pure Markov process. Both models give closed-form solutions and equations that are either mathematically equivalent or very closely matching. Both models yield equations for a number of important system performance metrics which include throughput, latency, packet loss, stability condition, CPU utilizations of interrupt handling and protocol processing, and CPU availability for user applications. To validate analysis, the results of both models are compared with reported experimental results as well as results of a discrete-event simulation.