Application Experience with Parallel Input/Output: Panda and the H3expresso Black Hole Simulation on the SP2
چکیده
Panda is a high-performance library for collective input and output of multidimensional array data on parallel platforms and networks of workstations. H3expresso is a numerical relativity code for modeling 3D spacetimes, including dynamic black hole simulations. This paper summarizes our experience in using Panda as the i/o component of H3expresso on the IBM SP2 at Cornell Theory Center. We show that Panda's high-level interface, which allows the user to request input or output of a set of arrays with a single command, is a good match for the needs of the H3expresso programmers. We describe three signiicant i/o performance issues that arise when using Panda with H3expresso: providing high performance despite the heterogeneity of the SP2, providing eecient oo-loading of output data given slow connections to the outside world, and satisfying the users' desire to dedicate as few nodes as possible to i/o. We explore the tradeoos between potential solutions to these problems, and present performance results for our approaches.