The implementation of effective quality of service (QoS) support for user services must be implemented end-to-end and through all layers of the OSI hierarchy at each network node. We consider the layer-2 design of an ATM switch, which is built as a scalable, 3-stage Clos switching network and discuss architectural choices for the implementation of QoS support on this switch. An ATM-like cell format is used to encapsulate and propagate traffic and resource management information through the switch. Call admission control, fast inter-stage cell transfers and priority-based queue arbitration are used as congestion control and resource management mechanisms. Simulation results are presented to show the effect of architectural choices (internal buffering) on QoS parameters.

This paper studies the non-blocking switching operation of generalized three-stage Clos (1953) switching networks in the multirate environment. The analysis presented determines the minimum number of the second-stage switches required for strictly non-blocking operation of such networks at call setup. Both the discrete and the continuous bandwidth cases are considered. For the discrete bandwidth case, sufficient and necessary conditions are derived. For the continuous bandwidth case, only sufficient conditions are given, which, in some cases, also constitute necessary conditions. The results given are, in some cases, generalizations of existing results, but they also include new results.

This paper studies the nonblocking switching operation of generalised 3-stage Clos (1953) switching networks in the multirate environment. The analysis presented in this paper determines the number of the second stage switches required for strictly nonblocking operation of such networks in the continuous bandwidth case. The results given are in some cases generalisation of known results, but they also include new results.

We present a high performance, scalable architecture for a video on demand (VoD) system, using a three stage Clos switch. Three stage Clos switching networks have been used in the past in multiprocessor interconnection networks as well as for ATM switching fabrics due to their attractive low latency and fault tolerance properties. The proposed VoD system mainly consists of such a three stage Clos switch, which scales to 1024 OC-3 inputs and 1024 OC-3 outputs, offering more than 132 Gbps of pure switching capacity. Video data is transported through the switching network using the ATM cell format and switching protocol, connection oriented network services and fast virtual circuit switching. Routing is centrally performed by means of intelligent nearly nonblocking control algorithms, developed for Clos networks, which increase the switch utilization and significantly reduce the blocking probability at call setup. The proposed architecture can manipulate up to several Terabytes of video data and simultaneously serve 22000 to 66000 out of a total of 262144 end users. Depending on the number of middle stage switches and the control algorithm used, the blocking probability at call setup can range between zero and 5e-4.

We study the semi-rearrangeably nonblocking (SRN) operation of asymmetrical three-stage Clos (1953) switching networks in the multirate environment. We develop a basic algorithm that balances the established connections among middle-stage switches by performing a small number of rearrangements per disconnection. For this algorithm, we first derive general conditions under which rearranging from a single middle-stage switch is sufficient to achieve SRN operation. In the most general case, however, a sequence of rearrangements from several middle-stage switches may be required for SRN operation. An algorithm to achieve this sequence of rearrangements is presented and its correctness is proved. The minimum resource requirements to achieve SRN operation, in terms of the number of middle-stage switches, are derived for various cases.

In this paper, we study the nonblocking operation of asymmetrical three-stage Clos networks. We consider a control algorithm for operating the. asymmetrical Clos networks in the nonblocking mode. We derive sufficient conditions under which these networks are nonblocking for this control algorithm. Further, we provide results on the number of faults tolerated by the algorithm in the nonblocking mode of operation. We next design a different control algorithm for nonblocking operation of asymmetrical Clos networks. Using simulation results, the new control algorithm is shown to perform better than the previous one in terms of network utilization, blocking probability and fault-tolerance.