. Analytical Comparison of Disha v.s. Avoidance-Based Wormhole Routing Algorithms A. Khonsari, A. F. Shahrabi, M. Ould-Khaoua Department of Computing Science University of Glasgow Glasgow, G12 8RZ, U.K. Email: {ak, Alireza, mohamed}@dcs.gla.ac.uk Abstract A critical requirement for any routing algorithm proposed for wormhole- switched k-ary n-cubes is to handle deadlock problem. Many adaptive routing algorithms using deadlock avoidance strategies have been dedicated some hardware resources especially to ensure deadlock freedom. Several recent studies have shown that deadlocks are quite rare, especially when enough routing freedom is provided. This consideration has motivated researchers to introduce fully adaptive routing algorithms with deadlock recovery. However, the performance of these new algorithms has been evaluated against deterministic routing only. In an effort to gain a deep understanding of the factors that affects routing performance, this paper compares the relative merits of Disha Concurrent [3] as an instance of a deadlock recovery algorithm to Duato’s adaptive routing as an example of deadlock avoidance algorithm. Our comparison also includes the well-known deterministic routing as it has been widely used as a performance test-bed in other similar studies in the past. While most existing network evaluation studies have been conducted using software simulation, the present study uses analytical models that have been shown to capture network performance with a good degree of accuracy. The results reveal that in most cases Disha routing exhibits the superior performance. 1 Introduction Networks belonging to the family of k-ary n-cubes have widely used in many practical multicomputers [2, 15, 18, 20, 21, 25] due to their desirable topological properties including ease of implementation, modularity, and recursive structure. A k-ary n-cube has N=k n nodes, arranged in n dimensions, with k nodes per dimension. Each node belongs to all n dimensions and is connected to two neighbours in each dimension. The two most popular instances of k-ary n-cubes are the hypercube (where k=2) and the torus (where n=2 or 3). The former has been used in multicomputers such as the SGI Origin 2000 [18], iPSC/2 [21] and iPSC860 [25] while the latter has been adopted in systems like the J-