Distributed control schemes for fast arbitration in large crossbar networks

In a large nonblocking crossbar switch, the controller often becomes a bottleneck in terms of both performance and reliability. We present a number of schemes for distributing the setup function among multiple controllers, thus improving both the performance and the reliability of the switch. The controllers are symmetric and operate in parallel. We present four distributed control schemes that provide a range of tradeoffs in controller complexity, speed, and hardware overhead for nonblocking operation. We derive a lower bound of N(1/spl minus/1/K) for the number of buses required for nonblocking operation of a crossbar switch with N ports and K controllers under certain constraints. We then describe a scheme that actually achieves this lower bound. Results from simulation indicate that the hardware overhead in terms of the extra buses needed is small for all the schemes if a small probability of blocking is acceptable.<
>