Previously, a quasi-static routing scheme called path switching was proposed for large-scale ATM packet switches, and a Clos network implementing this routing scheme is called a cross-path switch. In principle, a cross-path switch supports both multicast and multirate traffic and now we add the capacity of supporting per-connection quality of service (QoS) guarantees. In this paper we develop a model to guarantee per-connection QoS in cross-path switches. By characterizing the output traffic, we observe that due to the quasi-static nature of path switching, traffic always becomes more bursty after passing through the cross-path switch. This traffic distortion would greatly degrade performance guarantees at the successive switches along the path. In order to achieve better end-to-end performance guarantees, we propose two regulation schemes for the cross-path switch to reduce distortion. For comparison, their delay bounds are derived. Our results show that with the designed control strategies, both schemes can effectively preserve or even reduce the burstiness of individual traffic streams across the switch with optimal performance guarantees.

In this paper a novel class of network topologies known as the multi-dimensional shuffle-exchange network (MDSXN) is proposed. We show that the well-known de Bruijn graph and hypercube in fact both belong to the same class of graphs represented by MDSXN. The MDSXN is therefore the unification and generalization of the de Bruijn graph and hypercube. We show that members of MDSXN inherit the topological properties of both the de Bruijn graph and hypercube to a varying degree. This allows us to trade-off cost and performance effectively and construct networks which are most suitable for a particular purpose. Examples of applications of MDSXN include structure for switching and multicasting networks, optical network topology, and virtual topology for local and metropolitan area networks.

A generalized non-blocking copy network based on a broadcast Clos (1953) network is proposed. We show that if the set of input connection requests is ordered, the broadcast Clos network is non-blocking and route assignment can be done by using the rank of each connection request. Packet replications and routing are achieved by the generalized interval splitting algorithm. We show that the broadcast Clos network can be considered as the cascade combination of a reverse omega network and a broadcast omega network. The construction of copy network is therefore no longer limited to 2/spl times/2 switching elements. By recursively constructing the reverse omega and the omega networks using 2/spl times/2 switching elements, we show that the proposed copy network is the generalization of Lee's architecture (1988).

The implementation of ShuffleNets using a single star-coupler and slowly tunable transceivers implies maximum reconfigurability. By "maximum reconfigurability", the authors mean that any node can exchange its logical position in the network with another node by simply retuning their transceivers. This allows one to reconfigure the logical topology to adapt to the changing traffic pattern. In a multi-star network, however, one does not have the same degree of freedom for reconfiguration because one is not free to exchange the logical identities of an arbitrary pair of nodes. In the paper, the authors investigate the reconfigurability issue in the multi-star implementation of ShuffleNet. They show how one can attain maximum reconfigurability by properly choosing the system parameters. An algorithm for channels allocation is proposed to maximize reconfigurability under a particular configuration.>

ShuffleNet is one of the many architectures proposed for multihop lightwave networks. Its advantages include low mean-internodal distance and simple routing. Modular growth of ShuffleNets, however, is generally difficult and requires many hardware and software reconfigurations. The authors consider a multistar implementation of ShuffleNet and discuss how a (p,k) ShuffleNet can be expanded to a (p,k+1) ShuffleNet in modular phases, where each phase increases the number of nodes by only a small fraction and requires only minor hardware and software reconfigurations.>