Models of Parallel Applications with Large Computation and I/O Requirements

A fundamental Understanding of the Interplay of computation and I/O activities in parallel applications that manipulate huge amounts of data is critical to achieving good application performance, as well as to identifying principles for the design and management of parallel system software. In this paper we exploit a formal model of the behavior of large-scale parallel applications based on observations regarding the interaction of CPU and I/O activity in a representative set of scientific codes. The model can be applied at various levels of characterization granularity. Results from the model are in good agreement with measurement data from a set of representative I/O intensive codes. A capacity planning example is provided to illustrate one of the potential uses of our methodology in practice. The model is also used for performance prediction via a set of functional forms that effectively forecast the scalability of the computation and I/O components of an application. The complexity of the functional form adopted affects the accuracy of performance prediction.