Performance of the Cray T3D and Emerging Architectures on Canopy QCD Applications

TL;DR

This paper evaluates the performance of the Cray T3D system running Canopy QCD applications, comparing it to previous systems and analyzing factors affecting efficiency for grid-oriented scientific computing.

Contribution

It provides a detailed performance analysis of Canopy applications on the Cray T3D, including benchmarking results and insights into system limitations and scalability.

Findings

T3D achieves similar efficiency to ACPMAPS for QCD applications

Performance is limited by data access patterns and overheads

Eliminating Canopy overheads can nearly double efficiency

Abstract

The Cray T3D, an MIMD system with NUMA shared memory capabilities and in principle very low communications latency, can support the Canopy framework for grid-oriented applications. CANOPY has been ported to the T3D, with the intent of making it available to a spectrum of users. The performance of the T3D running Canopy has been benchmarked on five QCD applications extensively run on ACPMAPS at Fermilab, requiring a variety of data access patterns. The net performance and scaling behavior reveals an efficiency relative to peak Gflops almost identical to that achieved on ACPMAPS. Detailed studies of the major factors impacting performance are presented. Generalizations applying this analysis to the newly emerging crop of commercial systems reveal where their limitations will lie. On these applications, efficiencies of above 25\% are not to be expected; eliminating overheads due to Canopy will improve matters, but by less than a factor of two.