next up previous
Next: 2 The multi-Benes and Up: 1 Introduction Previous: 1.4 Our results

1.5 Our approach

The networks that we use to obtain these results are constructed by combining expanders and Benes networks in much the same way that expanders and butterflies are combined to form the multibutterfly networks described by Upfal [37]. We refer to these networks as multi-Benes networks. The nonblocking networks of Bassalygo and Pinsker [3] are similar. The details of the construction are provided in Section 2 of the paper.

The techniques in this paper can also be applied to bandwidth-limited switching networks such as fat-trees [21]. These networks may be more useful in the context of real telephone systems, where there are limitations on the number of calls based on the proximity of the calls (e.g., it is unlikely that everyone on the East Coast will call everyone on the West Coast at the same time).

The description and analysis of the path selection algorithm is divided into three sections. In Section 3, we prove that the multi-Benes network is a strict-sense nonblocking connector. A similar approach was used in [17] to show that the multibutterfly is capable of routing in the presence of many faulty nodes. Indeed, we can think of currently-used nodes as being faulty since they cannot be used to form new connections. Similarly, the algorithms we describe for routing in nonblocking networks can easily be extended to be highly tolerant to faults in the network. In Section 4, we describe an tex2html_wrap_inline1313 -bit-step algorithm for bit-serial routing in a multibutterfly. This algorithm relies on an unshared-neighbor property possessed by all highly-expanding graphs. By implementing this algorithm on the multi-Benes network and combining it with the methods of Section 3, we produce an algorithm that can handle many calls at the same time, independent of what calls have been made previously and what calls are currently connected.

In Section 5, we describe algorithms for handling multiparty calls, and situations where many inputs try to reach the same output simultaneously. Some of these algorithms rely on sorting circuits and are not as practical as those described in Section 4. We also show how to remove the distinction between terminals and non-terminals.



next up previous
Next: 2 The multi-Benes and Up: 1 Introduction Previous: 1.4 Our results



Bruce Maggs
Mon Jul 22 21:19:59 EDT 1996