Off-campus UMass Amherst users: To download dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.

Non-UMass Amherst users, please click the view more button below to purchase a copy of this dissertation from Proquest.

(Some titles may also be available free of charge in our Open Access Dissertation Collection, so please check there first.)

Cell graphs for managing communication in parallel computing

Bojana Obrenic, University of Massachusetts Amherst

Abstract

The hypothesis of this research is that current formal descriptions of parallel computation can benefit from introducing graph-oriented communication abstractions. We define a class of graphs, termed the cell graphs, which we employ to construct communication abstractions that contribute to the analysis and conceptual design of parallel computer architectures. We use the theory of network emulations to study communication in parallel computing. Emulations are algorithms for automatic translation of parallel programs designed for one interconnection network to another. The proposed cell-graph communication structure is conceived as a collection of specific emulations, such that original emulation problems are expressed as compositions of cell-graph emulations. Instead of solving the original emulation problem directly, the problem is decomposed into two independent parts: expressing the communication of the emulated graph in terms of the cell-graph primitives, and porting the cell-graph primitives to the emulator. In Chapter 1 we define the concept of cell graphs and derive their properties. In Chapter 2 cell graphs are used as an analysis tool; the results of this chapter include the first optimal algorithm for emulating arbitrary graphs having sublinear separators by shuffle-like interconnection networks. In Chapter 3 cell graphs are used as a design tool; this chapter is organized as a software design project the goal of which is to construct dynamically reconfigurable massively parallel architectures that can dynamically modify their datapath width and underlying graph structure, responding to changes in the parallelism grain inherent in applications.

Subject Area

Computer science

Recommended Citation

Obrenic, Bojana, "Cell graphs for managing communication in parallel computing" (1993). Doctoral Dissertations Available from Proquest. AAI9408324.
https://scholarworks.umass.edu/dissertations/AAI9408324

Share

COinS