Dynamic workload profiling and task allocation in packet processing systems

Computer networks require increasingly complex packet processing services on routers to adapt to new functionality, security, and performance requirements. Embedded multicore packet processing systems that can provide this capability are difficult to program and manage at runtime. We propose a novel way of representing processing tasks, obtaining runtime profiling information, and mapping tasks to processors. By duplicating processing tasks with heavy processing requirements, a more balanced workload can be obtained. The mapping algorithm considers that balance when assigning tasks to processors as well as the cost of inter-processor communication. Our evaluation results show that our approach can improve the system throughput by 2.39-2.89 times at a cost of 1.49-1.64 times higher inter-processor communication.