Scalable central-stage buffered Clos-network packet switches with QoS

In our previous work, we proposed a scalable packet switch architecture based on the central-stage buffered Clos-network (CBC). We analyzed the memory requirements for the CBC to emulate an output-queued (OQ) switch and left the corresponding scheduling algorithms unexplored. In this paper, we set out to find a practical algorithm to schedule packets in order for the CBC to emulate an OQ switch supporting quality of service (QoS). We observe that the CBC surprisingly extend the well-known Birkhoff-von Neumann input-queued switches, making it able to scale to large switches with many input/output ports. In particular, as far as we know, the most efficient scheduling algorithm for a Birkhoff-von Neumann switch has a time complexity of O(N^4.5), where N is the number of switch ports. We show in this paper that we can reduce it to O(N^2.25) by employing a multi-stage multi-layer switch implementation