Automated Design of Torus Networks

TL;DR

This paper introduces an algorithm for automatically designing torus network topologies for supercomputers, incorporating real-world cost and technical data to optimize hardware selection and compare with alternative network types.

Contribution

It presents a novel algorithm that integrates real equipment data into the design of torus networks, enabling cost-effective and optimized supercomputer network configurations.

Findings

The algorithm effectively incorporates real equipment costs.

It allows comparison between torus and fat-tree network costs.

Supports integrated hardware and network design decisions.

Abstract

This paper presents an algorithm to automatically design networks with torus topologies, such as ones widely used in large-scale supercomputers. The characteristic feature of our approach is that real life equipment prices and values of technical characteristics are used. As a result, we also have the opportunity to compare costs of torus and fat-tree networks. The algorithm is useful as a part of a bigger design procedure that selects optimal hardware of cluster supercomputer as a whole.