Title :
Supporting I/O-efficient scientific computation in TPIE
Author :
Vengroff, Darren Erik ; Vitter, Jeffrey Scott
Author_Institution :
Dept. of Comput. Sci., Brown Univ., Providence, RI, USA
Abstract :
We describe the facilities that TPIE provides to support scientific computation. TPIE, a transparent parallel I/O environment, is a system designed to assist programmers in implementing I/O-efficient computations. It supports I/O-efficient paradigms for problems from a variety of domains, including sorting and permuting, computational geometry, graph algorithms, and scientific computation. Our main focus in on the algorithmic issues faced in designing the portions of TPIE that support scientific computation. We consider four computing paradigms: scanning, sorting, sparse matrix methods, and dense matrix methods. Benchmark results using TPIE are also briefly described. References to more detailed discussions of these benchmarks, as well as the overall goals, philosophy, and structure of TPIE are provided
Keywords :
computational geometry; graph theory; input-output programs; natural sciences computing; parallel algorithms; programming environments; sorting; sparse matrices; I/O-efficient paradigms; I/O-efficient scientific computation support; TPIE; benchmark results; computational geometry; computing paradigms; dense matrix methods; graph algorithms; permuting; programmer assistance; scanning; sorting; sparse matrix methods; transparent parallel I/O environment; Algorithm design and analysis; Bandwidth; Computer science; Concurrent computing; Disk drives; Large-scale systems; Programming profession; Prototypes; Sorting; Sparse matrices;
Conference_Titel :
Parallel and Distributed Processing, 1995. Proceedings. Seventh IEEE Symposium on
Conference_Location :
San Antonio, TX
Print_ISBN :
0-81867195-5
DOI :
10.1109/SPDP.1995.530667