A vector-parallel implementation and statistical analysis of the bucket sort on a vector-parallel distributed memory system: Lessons learned in the integer sort NAS parallel benchmark

Conference · Fri Dec 01 04:00:00 EST 1995

OSTI ID:125626

Elton, B H; Miura, K ^[1]

---

1. Fujitsa America, Inc., San Jose, CA (United States)

The bucket sort is a very efficient algorithm for sorting integers when the range is known in advance. The major portion of this algorithm, however, is inherently sequential. We have selected, as a working example for this study, the variant of the bucket sort algorithm such as described in the NAS Parallel Benchmark suite of the NASA Ames Research Center Ill. In general, bow one implements the bucket sort algorithm on a distributed memory system depends on a variety of factors, including the communications bandwidth and latency, the type of network topology, the amount of memory per processor, the type of processor, and the size of the keys, i.e., the number of buckets. The VPP500 is a vector-parallel system with a distributed memory architecture, in which each processing element (PE) comprising a vector architecture. Attempting both efficient vectorization and parallelization of the bucket sort in a distributed memory system provides a backdrop for novel algorithm development.

• OSTI ID: 125626
• Report Number(s): CONF-950212--
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
ALGORITHMS
BENCHMARKS
DISTRIBUTED DATA PROCESSING
PARALLEL PROCESSING
TASK SCHEDULING