ON MODELING LOCAL PAGING BEHAVIOR FOR THE VAX/VMS SYSTEM
Abstract
Systems with paged virtual memories are difficult to model because the workload specification of a job depends on the collection of jobs running with it. Previous modeling studies have concentrated on systems with global paging algorithms such as IBM's VM/370 and MVS operating systems. This thesis develops a model of a paged virtual memory system with a local paging algorithm: the VMS operating system running on a VAX-11/780. Because many of the model's features do not easily yield to analytic solution, the model is based on discrete-event simulation. Process priority, preemptive queueing schemes, overlapped CPU-I/O processing by a single job, VMS quantum expirations, and I/O performed by Ancillary Control Processes are implemented in the model. Since VMS uses a shared page cache to improve paging performance, paging can be characterized by two parameters: page fault rate and paging I/O rate. A regression model is used to predict the paging I/O rate as a function of page fault rate, number of user jobs, and size of system memory. This regression model was incorporated into the system model. The system model was validated with a series of benchmark runs. Its predictive abilities were tested by comparing model performance with system performance as configuration parameters were varied. Parameters modified included number of user processes, size of memory, and user process resident set size. The model's prediction of the mean time needed to complete a user process was within ten percent of the correct value in most cases.
Degree
Ph.D.
Subject Area
Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.