From paracomputer to ultracomputer
The term paracomputer denotes an idealistic parallel processing machine in which each processor may access any cell within a global memory in unit time. Recently this model of parallel computation has come under attack because of its inability to deal with communications issues. The term ultracomputer denotes a more realistic parallel processing machine in which processors, each with some local memory, communicate with each other across some sparse interconnection network. In this thesis, the mapping of paracomputer algorithms onto ultracomputers is investigated in two ways. First algorithm-network configurations on ultracomputers are sought which retain the time, and when possible, the processor bounds of the paracomputer algorithm. The principal algorithm considered is matrix multiplication, and an algorithm-network configuration is provided that gives the strongest known upper bound for the matrix multiplication problem. The second approach is to develop upper and lower bounds for fundamental operations on the ultracomputers. Principal operations considered are those of collecting, broadcasting, and routing information. Lower bounds are derived for the broadcasting problems for both the ultracomputer and paracomputer models.
- Research Organization:
- Indiana Univ., Bloomington (USA)
- OSTI ID:
- 5401984
- Resource Relation:
- Other Information: Thesis (Ph. D.)
- Country of Publication:
- United States
- Language:
- English
Similar Records
NYU ultracomputer-designing an MIMD shared memory parallel computer
Sharing memory in distributed systems - methods and applications