Intelligent Prefetching and Caching for Scientific Datamining in the Middleware GAMine

A.C. Sodan and G. Hu (Canada)

Keywords

Prefetching, caching, middleware, adaptation, configuration

Abstract

We present a middleware (GAMine) for parallel scientific datamining that employs aggressive prefetching to hide communication and disk latency. GAMine can be configured for different caching/buffering policies, exploiting knowledge about the application characteristics. Furthermore, GAMine adapts its prefetch rate to the space being available for storing data and to the available throughput (which may change because of varying contention at the server or on the network).

Important Links:



Go Back