Shared memory switching + virtual output queuing: A robust and scalable switch

Abstract

A novel combination of shared memory switching and virtual output queuing yields an architecture that addresses the exponential growth of the Internet. Superior scalability is achieved using distributed schedulers compared to crossbar-based fabrics requiring a central scheduler. Performance is excellent and robust. We describe the system view, an actual implementation, performance simulations for several traffic classes, and QoS results using ban dwidth management.