Automatic parallelization of C programs using shared data-objects
Date
1990
Authors
Sinclair, Craig
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The method for exchanging data among processes, when developing programs for parallel execution, has a crucial impact on the design of the programs. In tightly coupled systems programs can directly reference common data through shared memory, whereas in loosely coupled systems programs typically exchange data through message passing. Architectures have recently emerged which integrate the shared variable and message passing paradigms. The most recent addition is the shared data-object model proposed by Bal and Tanenbaum.
This thesis presents an objected-oriented model for distributed programming based on the framework of the shared data-object model. Our model provides data replication through shared data-objects and process management through process objects. Sequential programs are partitioned automatically by means of a preprocessor into a set of program blocks that can be executed in parallel. Shared data-objects provide distributed flow control by blocking accesses that violate the ordering specified by the sequential program. A divide-and-conquer algorithm and a ray tracing rendering program are used to illustrate our methodology for distributed programming.