Performance Analysis on a CC-NUMA Prototype

Cache coherent non-uniform access machines (CC-NUMA) have been shown to be a promising paradigm for exploiting distributed execution. CC-NUMA systems can provide the improved execution performance that is typically associated when usiing parallel machines and are able to do so without the high programming costs commonly associated with parallel machines. This is because on a CC-NUMA machine, a single image of the memory system is provided. Past research on CC-NUMA machines has focused on modifications to the memory hierarchy, interconnect topology and memory consistency protocol, in an attempt to provide scalability. All of these areas are critical to achieving scalable performance. The research described here expands this focus to issues associated with operating system structures that can enable increased system scalability. We describe a hardware/software prototyping study that investigates how changes to the operating system of a commercial IBM AS/400 system can provide scalable performance when running transaction processing workloads. The project described was a joint effort between researchers at the IBM T.J. Watson Research Center and a team from the AS/400 development laboratory in Rochester, MN. This project is an example of a highly successful joint-project between development and research. This paper will describe various aspects of this project including the necessary changes that were made to the operating system to enable scalable performance, and the associated hardware and software performance tools developed to identify bottlenecks in the...