Services offered within the fiber-based broadband network may range from circuit emulations, where extremely low cell loss rates are essential, to high-speed variable rate services where a degree of cell loss can be tolerated but whose bandwidth requirements may be extremely bursty and unpredictable. Packet switches deployed within this network must be extremely robust to support such a broad spectrum of services. This paper presents a high-performance self-routing packet switch architecture, called Sunshine, that can support a wide range of services having diverse performance objectives and traffic characteristics. Sunshine is based on Batcher-banyan networks and achieves high performance by utilizing both internal and output queuing techniques within a single architecture. Parallel banyan routing networks provide access to output queues, allowing multiple cells to be delivered simultaneously to each destination. Excess cells, due to momentary output overloads, overflow into a shared recirculating queue to be delayed and resubmitted at dedicated switch inputs in the following packet slot. A priority structure supporting multiple levels ensures cells have access to each queue, based on priority, allowing multiple grades of service to coexist within the switch. Results based on extensive simulations, which modeled low-speed services by a random arrival process, circuit emulations by a periodic arrival process, and high-speed variable-rate services by a bursty process, show that this architecture can achieve the extremely low cell loss rates necessary for circuit emulations and is robust in a bursty environment. Next, an enhanced architecture is presented which allows the bandwidth from an arbitrary set of transmission links to be aggregated into trunk groups to create high bandwidth pipes. Trunk groups appear as a single logical port on the switch and can be used either to increase the efficiency of the switch in an extremely bursty environment, or to increase the access bandwidth for selected high bandwidth terminations. Simulation results show the effect of trunk grouping as a function of group size, peak access rate, and offered load assuming a datagram model where large blocks of information are fragmented for transport as a series of ATM cells. Using custom CMOS VLSI circuits and compact three-dimensional packaging, these architectures can support line rates in excess of 150 Mb/s.
