Gheorghe Almasi, Vernon Austel, Siddhartha Chatterjee,
John A. Gunnels, Fred G. Gustavson, and James C. Sexton
1
IBM, Ireland and the United States
email: sexton@maths.tcd.ie
Abstract:
The efficient execution of high-performance scientific codes is a primary goal of the BlueGene/L supercomputer. LINPACK, which involves the solution of a dense system of linear equations, is an established benchmark to determine the sustained performance of parallel computers.
In this paper, we detail the design and implementation of the LINPACK benchmark for BlueGene/L, covering the relevant aspects of the communications network, message-passing library, algorithmic design, performance modeling, and computational kernel construction. We conclude with an examination of the performance of the resulting code and an analysis of its potential performance on similar architectures.