Abstract

We propose a new scheme for multiplexing buffer space between flows contending for the same output port. In our scheme messages of the different flows are serviced in either Round-Robin or almost Round-Robin order, thus supporting fair queuing. Our scheme is both simple for hardware implementation and achieves high buffer space utilization (by taking advantage of statistical multiplexing between the flows). Therefore our scheme enjoys from both the simplicity of the hardware queue per flow scheme and the space efficiency of the linked-list based dynamic scheme, suggesting an attractive compromise between the two extremes. Keywords: Fair Queuing, Round-Robin, Deficit Round-Robin, Weighted Fair Queuing, Switch Architecture, Stochastic Multiplexing 1 Introduction We address the problem of sharing buffer space between backlogged flows that pass through a link. Since the traffic of the flows may be bursty, the link is occasionally congested, and backlogged messages are queued up. The sim...

Citations