Re-engineering statistical software for efficient parallel execution

✍ Scribed by K Murphy; M Clint; R.H Perrott

• Publisher: Elsevier Science
• Year: 1999
• Tongue: English
• Weight: 111 KB
• Volume: 31
• Category: Article
• ISSN: 0167-9473
• DOI: 10.1016/s0167-9473(99)00041-9

✦ Synopsis

Three basic operations frequently employed in the construction of statistical applications software are: vector orthogonalisation, solution of sets of linear equations and QR factorisation. In this paper the amenability to e cient parallel implementation on a Cray T3D computer of these operations is investigated: the ÿrst is addressed through the re-engineering of a serial implementation of the modiÿed Gram-Schmidt algorithm and the second and third through the re-engineering of a serial implementation of a composite algorithm for the solution of the equality-constrained linear least-squares problem. The e ectiveness of the parallel implementations is evaluated and some guidelines are o ered to those engaged in the construction of statistical applications intended for execution on distributed memory machines.