Performance Metrics for Ocean and Air Quality Models on Commodity Linux Platforms

This report examines performance of four models with two compilers on the Intel Pentium 4 Xeon processor using the PAPI performance event library to collect performance counter values. The intent is to identify performance metrics that indicate where performance inhibiting factors occur when the codes execute. Results for operations, instructions, cycles, cache, table lookaside buffer misses, branching instructions, and I/O operations are discussed in detail. Results support the thesis that codes optimized for vector register computer architectures also benefit from commodity cache-based architectures and the compilers that optimize for them. Consequences for Air Quality Models are analyzed by comparing results against two ocean models with a view to potential performance benefits on commodity architectures. Introduction This is a report on a project to evaluate industry standard fortran 90/95 compilers for IA-32 LinuxTM commodity platforms when applied to Air Quality Models (AQM). The goal is to determine the optimal performance and workload though-put achievable with commodity hardware. Only a few AQM’s have been successfully converted to OpenMP (CAMx, [1]) or MPI (CMAQ, [2]) and considerable work remains to be done on others. In exploring the potential for parallelism it has been interesting to discover the problems with serial performance on several AQM codes. For this reason we have searched for more precise metrics of performance as an aid to measuring progress in performance enhancement. The historical analogy is the programming environment on Cray architectures which enabled the development of performance attributes for either individual codes or workloads using hardware performance counters [3,4]. Since commodity processors also have performance counters, software interfaces, such as PAPI [5], may be used to read them. This study applied the PAPI library in understanding what delivered performance is for two AQM’s, ISCST3 [6] and AERMOD [7], and what the optimal achievable performance can be. For the latter, as a base-line, two Ocean models with good vector character have been included. These are used to measure the optimal performance to be expected on commodity hardware with available compiler technology. The intent in this report is to identify performance metrics that show where performance inhibition occurs on the hardware architecture, and not to so suggest what needs to be done