Data Placement Optimization Research Articles

THEMIS: COMPONENT DEPENDENCE METADATA IN ADAPTIVE PARALLEL APPLICATIONS

This paper describes THEMIS, a programming model and run-time library being designed to support cross-component performance optimization through explicit manipulation of the computation's iteration space at run-time. Each component is augmented with "component dependence metadata", which characterizes the constraints on its execution order, data distribution and memory access order. We show how this supports dynamic adaptation of each component to exploit the available resources, the context in which its operands are generated, and results are used, and the evolution of the problem instance. Using a computational fluid dynamics visualization example as motivation, we show how component dependence metadata provides a framework in which a number of interesting optimizations become possible. Examples include data placement optimization, loop fusion, tiling, memoization, checkpointing and incrementalization.

Synchronized data services over MPEG—2 digital television broadcast systems

AbstractUntil the present, the digital television (DTV) industry has focused providing high—quality basic video and audio services. With advancements in the underlying technologies, digital television is likely to evolve into a much more interactive medium, involving advanced data services and applications which are both integrated with and synchronized to the video/audio content of the broadcast. We present our approach for the implementation of these synchronized data services, with emphasis on the issues of content synchronization and the optimization of data placement in the broadcast stream.

Lower Bounds for Norms of Inverses of Interpolation Matrices for Radial Basis Functions

Interpolation of scattered data at distinct points xI,..., xn ∈ Rd by linear combinations of translates Φ(||x − xj||2) of a radial basis function Φ : R≥ 0 → R requires the solution of a linear system with the n by n distance matrix A ≔ (Φ(||xi − xj||2). Recent results of Ball, Narcowich and Ward, using Laplace transform methods, provide upper bounds for ||A−1||2, while Ball, Sivakumar, and Ward constructed examples with regularly spaced points to get special lower bounds. This paper proves general lower bounds by application of results of classical approximation theory. The bounds increase with the smoothness of Φ. In most cases, they leave no more than a factor of n−2 to be gained by optimization of data placement, starting from regularly distributed data. This follows from comparison with results of Ball, Baxter, Sivakumar, and Ward for points on scaled integer lattices and supports the hypothesis that regularly spaced data are near-optimal, as far as the condition of the matrix A is concerned.