Use of predictive performance modeling during large-scale system installation

We concern ourselves in this paper with one important application of predictive performance modeling - to validate the measured performance during system installation. In general, models can provide performance and scalability expectations of a system for a given workload. The application characteristics of the ASCI workload utilized in this paper is SAGE, a multidimensional, 3D, multi-material hydrodynamics code with adaptive mesh refinement. We review the salient features of an analytical model of this code that can be applied to predict its performance on a large class of large-scale parallel systems. We then utilize the model to validate system performance on a Compaq Alpha-server ES45 Supercomputing system being built at Los Alamos, and expected to grow to 30T peak performance in the next few years. We describe the methodology applied during system installation and upgrades to establish a baseline for the achievable 'real' performance of the system. We show that utilization of predictive performance models is also a powerful debugging tool