Concurrency control and scheduling for hard real-time systems
Abstract
Hard real-time systems (HRTS) are distinguished by explicit and stringent timing constraints. The problems of HRTS scheduling involve not only guaranteeing schedulability but also ensuring that shared data will not be corrupted. In the past, mutual exclusion techniques from concurrent programming (semaphores, monitors, etc.) have been directly applied in HRTS. However, performance may suffer, or incorrect results may be produced when synchronization is limited to mutual exclusion. Consistency problems arise when multiple processes read and update aggregate data, such as tables and buffers. Mutual exclusion at a very coarse level (whole data structures, or collections of data structures) may lead to unacceptable blocking, while mutual exclusion at a fine level (e.g., only for the duration of access to a single table element) may be insufficient for maintaining consistency. Much recent research in maintaining integrity of shared data while permitting concurrent access has been carried out in the context of database systems, where it is known as concurrency control. These techniques must be adapted and extended for the class of real-time systems in which tasks have stringent deadlines and at least some critical tasks must be analytically guaranteed to meet their deadlines (not with high probability but always). This dissertation adapts database concurrency control to scheduling HRTS and presents efficient scheduling and concurrency control protocols for HRTS. Our approach to adapting concurrency control techniques employs semantic information that is necessarily available at design time for the class of HRTS requiring analytic guarantees of schedulability. For example, we have adapted a multiversion integrated concurrency control protocol and a mixed locking and abort protocol which use knowledge of task periods and data access sets to reduce blocking and enable predictable, low-overhead implementation with simple data structures.
Degree
Ph.D.
Advisors
Young, Purdue University.
Subject Area
Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.