SERA: Service Ratio Based Preemption for Minimizing Throughput Loss in Networks

Resource allocation is a fundamental problem in the connection oriented network environment. One key problem is to address the situation when the requested bandwidth exceeds the capacity of the network link. A decision has to be made whether to reject the new request or preempt any existing connections in order to admit the new request. In the event that preemption is triggered, multiple existing connections may need to be preempted in order to free sufficient resources for the new request. The time invested and work done for existing connections may be lost. This will affect the network revenue as well as its overall throughput. In this paper, we present a decentralized service ratio arbitrated (SERA) preemption strategy to minimize any loss of throughput. We propose to use the ratio of the connection´s remaining holding time and its elapsed time as the suitable preemption criteria. It will be better to preempt a connection that has just started since the work and time invested so far are minimal. Finally, simulation and analytical results confirm that the preemption strategy can significantly reduce the degradation that would otherwise occur for the throughput.