Approximate-solution techniques for queueing network models of concurrent processing and other nonproduct form problems

Thesis/Dissertation · Thu Jan 01 04:00:00 EST 1987

OSTI ID:6138080

Almeida, V A.F.

Concurrency is a central issue in current and future computer systems. In the last few years, several computers with advanced architectures have been announced. All of them employ some form of concurrency to achieve higher performance. This dissertation provides an improved trade-off between accuracy and simplicity in modeling programs with internal concurrency. Several different synchronization mechanisms are analyzed. A reduction technique is developed that analyzes the performance of parallel-processing systems and that preserves the simplicity of standard queueing network-theory techniques (e.g., the mean value analysis algorithm). The technique provides quite accurate approximations on the performance of such systems. Extensions related to the applicability of the reduction technique are also presented. The technique successfully applies to other classical non product form problems (e.g., simultaneous resource possession and routing to the shortest queue). The significance of this research stems from the need for models that yield insights into the behavior of parallel-processing systems.

• Research Organization: Vanderbilt Univ., Nashville, TN (USA)
• OSTI ID: 6138080
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
990210* -- Supercomputers-- (1987-1989)
COMPUTER ARCHITECTURE
COMPUTER NETWORKS
NETWORK ANALYSIS
PARALLEL PROCESSING
PERFORMANCE
PROGRAMMING
QUEUES
SYNCHRONIZATION