K.J. Barker and D.J. Kerbyson (USA)
Performance modeling, performance analysis, and large-scale applications
In this work we present a detailed analytical performance model of the large-scale parallel application HYCOM, the Hybrid Coordinate Ocean Model. The performance model is developed from analyzing the activities contained within the code. It is parameterized in terms of both application characteristics (including problem sizes, time-steps, and application phases) and system characteristics (including single processor performance, computational node size, network latency, and network bandwidth). Through a validation stage it is shown that the model is reasonably accurate with a typical prediction error of 10% when compared with actual run-times across three parallel systems and when using three different input decks. Once developed and validated the model may be used to explore performance scenarios, consisting of application and system combinations, for which it may not be possible to directly measure. One study is presented here which is used to provide insight into the achievable performance on a possible future (2010 timeframe) system. In addition we use the performance model to show the sensitivity of HYCOM’s run-time to the performance of the individual sub-systems: processor speed, network latency, and network bandwidth. These studies illustrate the power of performance modeling once an accurate model has been developed.
Important Links:
Go Back
