Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

A programming paradigm for distributed-memory computers

Conference ·
OSTI ID:10176962
One of the problems that arises in programming a multicomputer is the effective partitioning of the work into tasks and the assigning of those tasks to the processors. We will present a programming paradigm as a method of structuring the algorithms, allowing us to identify and separate programming phases. This paradigm allows us to develop a general software tool for dynamically allocating work to the processors while hiding many details of load balancing. Our paradigm consists of four phases: partitioning the work, mapping the tasks into the processors connected by some virtual topology, running the application program, and embedding the virtual architecture into the actual machine. The paradigm applies to diverse problems and to a variety of multiprocessors without significant reprograming. We will use the multisection method for computing eigenvalues to show how this tool works.
Research Organization:
Colorado Univ., Boulder, CO (United States). Dept. of Computer Science
Sponsoring Organization:
USDOE, Washington, DC (United States); National Science Foundation, Washington, DC (United States)
DOE Contract Number:
FG02-92ER25122
OSTI ID:
10176962
Report Number(s):
DOE/ER/25122--5; CONF-930331--13; ON: DE93018336; CNN: Grant CCR-9109785
Country of Publication:
United States
Language:
English

Similar Records

Paradigm; A highly scalable shared-memory multicomputer architecture
Journal Article · Thu Jan 31 23:00:00 EST 1991 · Computer; (USA) · OSTI ID:5533695

Optimal eigenvalue computation on distributed-memory MIMD multiprocessors
Conference · Thu Oct 01 00:00:00 EDT 1992 · OSTI ID:10177289

Program partitioning for NUMA multiprocessor computer systems. [Nonuniform memory access]
Journal Article · Sun Oct 31 23:00:00 EST 1993 · Journal of Parallel and Distributed Computing; (United States) · OSTI ID:5703692