Dynamic processor self-scheduling for general parallel nested loops
In this paper we present a completely dynamic processor self-scheduling approach for general nested loops. General nested loops contain both parallel and serial loops and the execution time of their iterations can vary widely. In our scheme, an instance of an innermost loop is considered to be a basic unit in a precedence graph. An instance is said to be active if the instance is ready to be executed. Completion of an instance will activate instance of other innermost loops. The basic concept in our scheme is to keep all processors busy as long as there exists any active instance. By effectively using synchronization primitive provided in a parallel processing system, the overhead of our approach can be significantly reduced. We also present the data structure needed to represent a general nested loop in our self-scheduling scheme. 19 refs., 8 figs.
- Research Organization:
- Wichita State Univ., KS (USA). Dept. of Computer Science; Illinois Univ., Urbana (USA). Center for Supercomputing Research and Development
- DOE Contract Number:
- FG02-85ER25001
- OSTI ID:
- 5642684
- Report Number(s):
- DOE/ER/25001-65; CONF-870831-14; ON: DE88003572
- Country of Publication:
- United States
- Language:
- English
Similar Records
Deadlock prevention in processor self-scheduling for parallel nested loops
Dynamic processor self-scheduling for general parallel nested loops