Approximations for second moments of performance parameters in distributed database networks

Thesis/Dissertation · Sun Jan 01 04:00:00 EST 1989

OSTI ID:6090989

Ronen, Y

Distributed database management systems permit users, from local and remote sites, to access and modify stored data concurrently. Updates made to the database must be carefully controlled to maintain data integrity constraints. Concurrency control algorithms synchronize transaction executions to ensure database integrity. As a result of obeying concurrency control rules, transaction executions incur an overhead. In distributed database systems, the overhead can be significant because it reduces parallelism and increases local work load thus reducing performance. Most analytical models for concurrency control performance evaluation have been developed for centralized systems. The author suggests analytical models for studying the second moments of performance parameters in distributed database systems controlled by locking algorithms. Synchronization is a key feature of concurrency control locking algorithms. His study shows that synchronization causes probability distribution functions with clustered probability masses. This type of probability function has very high variance. The conclusion of his study is that such systems must be analyzed using second moments, otherwise the results might be misleading. This thesis is an extension of the Mean Value Approximation Model (MVAM), that has been developed at Drexel CCINR. However, MVAM approximates only the first moments of the performance parameters. His modeling approach is based on decomposing the network into solvable portions and approximating the behavior of each portion as a G/M/1 queueing system. A virtual queueing network is proposed to approximate the parameters of the distribution functions of the arrival processes to the G/M/1 nodes. The virtual queueing network incorporates the concurrency control protocols into specially defined waiting nodes. A simulator is used to validate the suggested model. There is good agreement between the analytical model and the simulation results.

• Research Organization: Drexel Univ., Philadelphia, PA (United States)
• OSTI ID: 6090989
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
990200* -- Mathematics & Computers
ALGORITHMS
COMPUTER NETWORKS
COMPUTERIZED SIMULATION
DATA BASE MANAGEMENT
DATA PROCESSING
DISTRIBUTED DATA PROCESSING
INFORMATION SYSTEMS
INFORMATION VALIDATION
MANAGEMENT
MATHEMATICAL LOGIC
PARALLEL PROCESSING
PERFORMANCE
PROCESSING
PROGRAMMING
SIMULATION
SYNCHRONIZATION
TESTING
VALIDATION