Resource management using dynamical load prediction and multiprocessor cooperation

摘要

Resource management is a complicated problem in multiprocessor system. When tasks with real-time characteristic are scheduled on processor resource constraints such as CPU and memory have to be met, however, it is usually difficult for a system to ensure load balancing and resource constraints simultaneity. In order to handle this problem lots of algorithms have been brought forward. In this paper, a load balancing policy is proposed to manage and integrate processing resource of multiprocessor system which is used to send packets of packets flows on time. The policy dynamically predicts the wait time before a packet to be processed based on the load of the processor and the timestamp of the packet, and decides the migration of packet between processors to guarantee that the packets can be processed before deadline. This policy can not only increase the multiprocessor utilization but also enhance the capacity of the system against the jitter of load by packets migration between the processors as well. We use simulation experiment to show that the policy can reduce the possibility of packets delay and increase the concurrency of flows that the system can serve.1