Abstract

In a network management system, software sensors (agents) collect system information and notify the central control system utilizing UDP/IP packets. When the network suffers a failure, the packet receiver encounters a flood of messages from a large number of agents. In such a situation, the receiver has to capture all these messages and process them in real-time. The thesis addresses this problem about how to design and develop a receiver that can capture large volume of packets while consuming minimum CPU resources. Linux network stack, contemporary related research work and some high capacity packet capturing solutions like NAPI and PFRING were studied to expose their limitations. Issues that limit the performance of these architectures have been identified and addressed in this thesis. Based on this analysis, a few design principles have been identified which can be applied to design and implement a high capacity, efficient packet capturing solution