Climate change studies need numerical models of the earth-atmosphere system to be integrated for extended periods of time (typically, climate models are run for several decades to study global change). Coupled ocean-atmosphere models need to be integrated for much longer periods (100 simulated years). Such simulations, the need for higher resolutions, and the increasing sophistication of physical parameterizations will require extensive computational resources. Scalable parallel computers will provide the increase in computational speed necessary for longer runs at higher model resolutions, but are subject to inefficiency in the form of computational load imbalance.
In this study we discuss the variation of computational load in physics modules of a global climate model and the load imbalances that result when the code is implemented on a massively parallel computer. The study was conducted using PCCM2, a parallel implementation of the NCAR Community Climate model (CCM2) running on the Intel Touchstone DELTA computer.