Resource aware packet scheduling for multi-resource in-network nodes

Abstract

Network has been widely reported as a bottleneck resource for amounts of data transferring in cloud and big data computing frameworks. In this paper, we focus on the heterogeneous flows problem that various flows consuming different amount of multiple resources. We develop a Resource Aware Queuing (RAQ) system to classify various flows into different categories according to their resource requirement amounts, and then schedule their processing order based on the switch multiple resource utilization. Based on the resource aware scheduling we also modify the packets dropping strategy so that the sending rates of different flows are also correspondingly modulated. We evaluate RAQ performance via simulation experiments and the results demonstrate that RAQ can improve network resources utilization and throughput without imposing serious negative impact on packets delays.