Implementing traffic shaping and link scheduling on a high-performance server

This paper examines the implementation of traffic shaping and link scheduling mechanisms on an end-host server. We first compare various implementation strategies, and then present a network interface architecture with dedicated support for traffic shaping and link scheduling for use in end-host servers. This results in significant load reduction on the server CPU, as shaping and scheduling tasks can execute concurrently on the network interface. This allows the server to provide very fine-grain link multiplexing, and consequently more diverse set of QoS guarantees for a large number of flows. We present two alternative implementations of our architecture. The first uses custom hardware while the second is implemented as a software component running on a dedicated processor on the network interface.