A Flow-Based Parallel Packet Processing Approach for Core Affinity and Core Scalability

Abstract

Existing high-speed parallel packet processing approaches lack in supporting core affinity and scalability. The core affinity is important to improve the packet processing performance for the flow, and the core scalability is adjusting the number of cores of a network application, that is required to achieve parallel packet processing efficiency, scaling based on the incoming network traffic. In this paper, we propose an approach to support flow core affinity and core scalability simultaneously. We allow multiple cores to handle packets in parallel while ensuring the flow core affinity by using a flow-to-core table. We also allow one packet-receive queue for one core and monitoring traffic workload, we automatically adjust the number of packet-receive queues and cores according to the traffic workload for core scalability. Through experiments, we examine the feasibility of the proposed approach.