Implementing functional languages to exploit locality
In the quest for high performance, no obstacle has been as persistent or unyielding as memory latency. It was hoped that dataflow's fine-grain asynchronous model of execution might defeat the memory latency problem. Unable to realize efficient fine-grain systems, the dataflow community is now studying medium-grain and coarse-grain implementations which, like conventional execution models, suffer the effects of memory latency. In this paper, we describe a functional language implementation that automatically exploits locality on cache-coherent multiprocessors. Our system achieves performance improvements reaching 20% for some programs. This study lends further support to the superiority of the functional paradigm for parallel processing. 11 refs., 9 figs.
- Research Organization:
- Lawrence Livermore National Lab., CA (United States)
- Sponsoring Organization:
- DOE; USDOE, Washington, DC (United States)
- DOE Contract Number:
- W-7405-ENG-48
- OSTI ID:
- 5241470
- Report Number(s):
- UCRL-JC-107491; CONF-911230--2; ON: DE91015121
- Country of Publication:
- United States
- Language:
- English
Similar Records
A prototype functional language implementation for hierarchical-memory architectures
Single-assignment language in a distributed-memory multiprocessor. Research report