Improving the performance scalability of the community atmosphere model

The Community Atmosphere Model (CAM), which serves as the atmosphere component of the Community Climate System Model (CCSM), is currently the most computationally expensive CCSM component in typical configurations. Improving performance scalability in CAM has been a challenge, due largely to algorithmic restrictions necessitated by the polar singularities in its latitude-longitude computational grid. Nevertheless, through a combination of exploiting additional parallelism, implementing improved communication protocols, and eliminating scalability bottlenecks, we have been able to more than double the maximum throughput of CAM on production platforms. We describe these improvements and present results on the Cray XT4/XT5, IBM BG/P, and an Opteron/Infiniband cluster. This improved performance will enable the CCSM research community to use its computing resources more effectively, allowing additional and more computationally expensive experiments to be run in support of the upcoming Intergovernmental Panel on Climate Change (IPCC) fifth assessment.