doi:10.1016/j.comcom.2003.08.011
Copyright © 2003 Elsevier B.V. All rights reserved.
Performance analysis of efficient multipath crossbars
Department of Electrical Engineering, San Jose State University, One Washington Square, San Jose, CA 95192, USA
Received 9 August 2002;
Revised 8 August 2003;
accepted 13 August 2003.
Available online 2 October 2003.
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
Abstract
This paper presents a comprehensive analysis of high-performance crossbars where crossbars can be asymmetrical in size, buffered at all crosspoints, and capable of multipath routing. A crossbar, as a building block of switching networks, with the mentioned features in its structure can demonstrate fault-tolerant property. This property is a major factor for crossbars to play the role of high-performance entity in modern computer and communication systems. A generic condition of n×k crossbar with n inputs (rows), k columns, and n outputs (rows) under which a crossbar becomes multipath is examined. In a multipath crossbar a packet can be routed through k different paths to an output. The multipath feature ensures the property of fault-tolerance and provides a faster congestion resolution especially under the multicast traffic. This paper describes a two-dimensional Markov chain associated with the queueing model, and presents the performance evaluation of crossbars.
Author Keywords: Author Keywords: Switching network; Fault-tolerant property; Crossbars
Fig. 1. A multipath crossbar (n=k=4) and its routing.
Fig. 2. A blocking model for the multipath crossbar.
Fig. 3. A blocking probability of the multipath crossbar excluding queueing impact.
Fig. 4. A queueing model for the multipath crossbar.
Fig. 5. Signaling in the multipath crossbar.
Fig. 6. Throughput (T) of a 16-port network when offered load (ρ) and number of columns (k) vary.
Fig. 7. Delay (D) in a 16-port network when offered load (ρ) and number of columns (k) vary.
Fig. 8. Throughput (T) when the network dimension (n) and the number of columns (k) simultaneously increase.
Fig. 9. Delay (D) when the network dimension (n) and the number of columns (k) simultaneously increase.
Fig. 10. Throughput (T) versus the number of columns (k) where the number of buffers per crosspoint (β) is 1 or 2.
Fig. 11. Complexity of a generic crossbar per port versus the network size (n) and the number of buffers at the crosspoint (β).
Table 1. Comparison of simulation and analysis on the throughput, T, versus offered load, ρ, when the number of buffers per crosspoint β=1, network size n=16

Table 2. Comparison of simulation and analysis on the packet delay, D, versus offered load, ρ, when the number of buffers per crosspoint β=1, network size n=16
