Abstract KeLP

1 Abstract We describe a set of abstractions for managing elaborate irregular data structures that communicate within Cartesian subspaces. These abstractions have been implemented as part of KeLPTools, a joint eeort involving research groups at UCSD and LBNL. We discuss the application of KeLPTools to a challenging problem involving structured adaptive mesh reenement, and discuss progress to date. 2 Introduction Irregular representations play an important role in solving many challenging scien-tiic problems, for example,: adaptive reenement, front tracking, shock capturing, and porous media ow. In many cases, the communication patterns may be described in terms of an underlying physical lattice embedded in Z n , where n is the dimensionality of the space. In previous work, Baden and co-workers 2, 3, 1, 4] have explored communication abstractions for irregular blocked data. The resultant infrastructure, called KeLP , has been applied to a variety of applications including multi-block methods for subsurface ows 6], Direct Numerical Simulation of turbulencee5], and struc-tured adaptive reenement for ab-initio molecular dynamics 13]. A distinguishing feature of KeLP is the provision for geometric meta data in expressing data motion and data decomposition. Variants of this model, explored in