Performance Analysis of a Network Sensor's Packet Processing System using Generalized Stochastic Petri Nets

Abstract

This paper describes a modeling based on Generalized Stochastic Petri Nets (GSPN) to analyze the performance of a network probing node in terms of throughput. The probing node is part of a distributed monitoring system. In this environment, the use of multiprocessor and multicore systems, as well as the parallelization of applications, is aimed at improving the node performance. Petri nets allow not only to represent the parallelization feature, but also to include the main events identified in the system: the packet arrival and a two-stage processing. The two-stage processing consists of a first stage in which packet capturing functionalities are performed, and a second stage in which a deeper packet treatment is performed. In addition, the Petri net model can reproduce a shared buffer control mechanism to ensure the integrity of the data. After detailing all the model components, the verification and validation of the model are done by using a simulation tool. With this model, it is expected to estimate the efficiency of the probing node early in the design and development stages.