Performance estimation of distributed computer systems

Thesis/Dissertation · Fri Jan 01 04:00:00 UTC 1988

OSTI ID:5702068

Morse, J A

Many computer systems encountered in the real world are best modeled as sets of independent sequential processes whose interaction is limited to synchronized message exchange. Such communicating sequential processes are often non-deterministic - their behavior can only be described in probabilistic terms. This thesis presents analytic methods for predicting the performance of these types of systems. The method uses an adaptation of C. A. R. Hoare's CSP language as a specification language. A Petri net is constructed from the CSP specification to capture the concurrency and synchronization constraints of the system. By doing reachability analysis on the Petri net, a probabilistic grammar is derived that describes the possible sequences of messages that could arise from execution of the system under study. Analysis of the probabilistic grammar predicts the mean and standard deviation of the total message traffic generated, and also predicts the mean number of occurrences of each message type. Several unique aspects of this approach are described, including transformation and simplification techniques that help avoid the state explosion problem inherent in this type of modeling.

• Research Organization: Connecticut Univ., Storrs, CT (USA)
• OSTI ID: 5702068
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
990300* -- Information Handling
COMPUTER NETWORKS
DATA PROCESSING
DISTRIBUTED DATA PROCESSING
FUNCTIONAL MODELS
PERFORMANCE
PROCESSING
PROGRAMMING LANGUAGES