Exploiting Workload Parallelism for Performance and Power Optimization

To achieve application performance, future power-limited design constraints impose new rules on the scaling for applications. Where in the past, application performance was the domain of delivering peak performance and translating peak performance into actual application performance, the challenge for future supercomputing applications will depend on delivering the best performance for a given budget.

To determine the impact of application scaling parameters to deliver performance for a given power budget, we analyze three applications These applications include both strong and weak scaling applications, as well as a study of the effect of input sets on power/performance scaling characteristics of an application. Specifically, we study the NAMD, UMT2K and WRF codes using a Blue Gene/L system as the target platform. We find that even for strong scaling problems, Blue Gene/L systems can deliver superior performance scaling and deliver significant power/performance efficiency.

Application benchmark power/performance scaling for the voltage-invariant energy X delay² power/performance metric demonstrates that choosing a power-efficient 700MHz embedded PowerPC processor core and relying on application parallelism was the right decision to build a powerful, and power/performance efficient system.