Load Scheduling for Flow-based Packet Processing on Multi-core Network Processors

Load scheduling is critical to the performance of parallel processing network devices. With the rapid development of multi-core technology, efficient load scheduling scheme optimized for multi-core network processors becomes increasingly important and motivates intensive research today. In this paper, we study the relationship between two canonical scheduling schemes, packet-level scheduler and flow-level scheduler, and find out that scheduling at flow-slice level can further exploit parallelism while preserving per-flow packet-order. An adaptive load scheduling scheme at flow-slice level is proposed and evaluated. The experiment results show that this scheme can achieve better balance of workload than that of flow-level scheme, while keeping high cache utilization rate in typical system configurations.