Computer Interconnection Networks Research Articles

Prophet: automated scheduling of SPMD programs in workstation networks

Obtaining efficient execution of parallel programs in workstation networks is a difficult problem for the user. Unlike dedicated parallel computer resources, network resources are shared, heterogeneous, vary in availability, and offer communication performance that is still an order of magnitude slower than parallel computer interconnection networks. Prophet, a system that automatically schedules data parallel SPMD programs in workstation networks for the user, has been developed. Prophet uses application and resource information to select the appropriate type and number of workstations, divide the application into component tasks and data across these workstations, and assign tasks to workstations. This system has been integrated into the Mentat parallel processing system developed at the University of Virginia. A suite of scientific Mentat applications has been scheduled using Prophet on a heterogeneous workstation network. The results are promising and demonstrate that scheduling SPMD applications can be automated with good performance. Copyright © 1999 John Wiley & Sons, Ltd.

Embeddings of Butterflies into Hypermeshes

Hypermeshes have been given much attention as a versatile interconnection network of parallel computers. A hypermesh is obtained from a mesh by replacing each linear connection with a hyperedge. In this paper, we show how to embed a butterfly or multiple copies of a butterfly into a hypermesh. First, a butterfly B(s) of (s + 1)2s nodes is embedded into a 2s × X hypermesh where X = 2⌊ log 2 s ⌋+ 1. Second, the butterfly B(s) is embedded into a square hypermesh. Third, multiple copies of the butterfly B(s) are embedded into a hypermesh of variable aspect ratio. The efficiency of these embeddings is measured by alignment cost, congestion, and expansion. The alignment cost of all of these embeddings is optimal. The congestion of the first and third embedding is optimal. The expansion of the first and third embedding is one if s = 2k - 1 for some integer k, otherwise, less than two. The expansion of the second embedding is 2 + ∊ (s) where ∊(s) = (2 log (s + 1) + 2)/(s + 1).

THE CUBE-OF-RINGS INTERCONNECTION NETWORK

We present a family of interconnection networks named the Cube-Of-Rings (COR) networks along with their basic graph-theoretic properties. Aspects of group graph theory are used to show the COR networks are symmetric and optimally fault tolerant. We present a closed-form expression of the diameter and optimal one-to-one routing algorithm for any member of the COR family. We also discuss the suitability of the COR networks as the interconnection network of scalable parallel computers.

Routing in optical multistage interconnection networks: a neural network solution

There has been much interest in using optics to implement computer interconnection networks. However, there has been little discussion of any renting methodologies besides those already used in electronics. In this paper, a neural network routing methodology is proposed that can generate control bits for a broad range of optical multistage interconnection networks (OMIN's). Though we present no optical implementation of this methodology, we illustrate its control for an optical interconnection network. These OMIN's can be used as communication media for distributed computing systems. The routing methodology makes use of an artificial neural network (ANN) that functions as a parallel computer for generating the routes. The neural network routing scheme can be applied to electrical as well as optical interconnection networks. However, since the ANN can be implemented using optics, this routing approach is especially appealing for an optical computing environment. Although the ANN does not always generate the best solution, the parallel nature of the ANN computation may make this routing scheme faster than conventional routing approaches, especially for OMIN's that have an irregular structure. Furthermore, the ANN router is fault-tolerant. Results are shown for generating routes in a 16/spl times/16, 3-stage OMIN.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

On Embeddings in Cycles

We prove exact results on dilations in cycles for important parallel computer interconnection networks such as complete trees, hypercubes, and 2- and 3- dimensional meshes. Moreover, we show that trees, X -trees, n -dimensional meshes, pyramids, and trees of meshes have the same dilations both in the path and in the cycle.

The Embedding of Meshes and Trees into Degree Four Chordal Ring Networks

In this paper a particular computer interconnection network, the symmetric chordal ring network of degree four, is presented, and the mapping of meshes and binary trees onto chordal ring networks is analysed. Expressions for the network diameter (the maximum distance a message must travel between any pair of computers) and the mean inter-computer distance are derived for a sub-set of chordal ring networks. Such networks incorporate the maximum number of computers for a given diameter, and have a communications cost, measured either as network diameter or as the mean internode distance, of O(\/N). While these networks provide attractive properties for mesh-based applications on small- and medium-sized multicomputer systems, binary trees are restricted to five levels (31 nodes).

Static interconnection network extensibility based on marginal performance/cost analysis

The paper examines the trade-off between cost and performance of a hypercube based computer interconnection network founded on a detailed cost analysis model. A cost function is proposed which considers the economies of scale of bandwidth available through optical fibre communication and the nonincremental increase in node complexity resulting from variations in network size. When considering topology choice, an important consideration is expansion. Any comparison of expansion should be done through marginal analysis: weighing the additional benefits (performance) against the additional cost resulting from expansion. To demonstrate this technique, four interconnection networks are considered: boolean n-cube, nearest neighbour mesh hypercube, generalised hypercube, and spanning multiaccess channel hypercube. Systems with equalised performance are studied, investigating the variations in cost. This permits the investigation of the trade-off of performance/cost between high capacity communication channels and increased number of channels and degree.