Content area
Abstract
Run-time monitoring has been applied in software-intensive systems to detect run-time constraint violations and trigger system recovery actions. Uncontrolled monitoring activities may, however, delay detection of a violation for an unbounded time and, worse, affect the original system's schedulability. In this dissertation, we introduce the concept of predictable run-time monitoring, which demands a bound on detection latency while ensuring temporal non-interference by the monitoring process. We present off-line analysis techniques that can 1) predict the maximum detection latency and the maximum (event) queue length with fixed-priority scheduling on uniprocessor systems, and 2) control the processor resource for monitor tasks and target tasks.
To support run-time monitoring on rapidly growing multi-core applications, we extend the uniprocessor Deferrable Server algorithm to the Synchronized Deferrable Server algorithm for fixed-priority multi-core or multiprocessor systems. The Synchronized Deferrable Server algorithm combines partitioned and global multiprocessor scheduling in one system, and can efficiently utilize processor bandwidth for run-time monitoring. We present off-line analysis techniques that can bound from above the maximum detection latency of a monitor task scheduled by the Synchronized Deferrable Server algorithm. Given a fixed amount of processor bandwidth for a set of Synchronized Deferrable Servers running on different cores, we compare three different schemes of allocating bandwidth to each of these servers. We show that evenly allocating bandwidth is "better" than other allocation schemes in terms of a task set's schedulability.
In addition to the predictable run-time monitoring theories, we implement the deferrable server algorithm as a Linux kernel module that provides hard real-time performance based on Xenoma, which cooperates with the Linux kernel to provide a real-time execution environment. This kernel module allows the implementation of predictable run-time monitoring on hard real-time systems.
