Configuring a Software Router by the Erlang-<inline-formula> <tex-math notation="LaTeX">$k$ </tex-math> </inline-formula>-Based Packet Latency Prediction

Abstract

Characterizing a virtual network function performance is one of the most challenging issues in softwarized network function deployment due to different configurations leading to a different performance. Since there has been no clear solution on how to configure a software router that provides the minimum average packet latency, we address this problem by proposing a configuration framework for a software router. First, we advance the performance characterization of a software router by conducting a packet latency prediction model based on the Erlang- $k$ distribution. Motivated by the analysis of the packet latency distribution, the Erlang- $k$ distribution is chosen as a basis of the packet latency prediction model. Our prediction model requires the measurement of only two different configurations, i.e., one and two receiving queues of a network interface card, to predict the average packet latency of all configurations. We cross-verify the accuracy of our prediction model with the measured data. Second, we adopt our prediction model in the configuration selection (CS) algorithm for searching which configuration yields the minimum of average packet latency. The validity of the prediction-based CS algorithm is confirmed by comparing to the measurement-based one. Finally, we present the CS-based measurement system interworking with network function virtualization-management and orchestration for advances in network management.