Dynamic Memory Management In A Resource-Constrained Real-Time Utility Accrual Environment

This proposal presents MSA, a uniprocessor, single address space resource scheduling algorithm. MSA addresses the open research problem of explicit dynamic memory management in a resource-constrained real-time utility accrual environment. As an overload scheduler, the algorithm ensures graceful performance degradation as defined by a set of optimality criteria. MSA extends the scope of resource scheduling to include main memory, and dynamically evaluates the system-wide effect on utility of biased memory management during memory overload conditions. Furthermore, preliminary work has been performed to enable MSA to perform scheduling in an environment where task/resource dependencies can dynamically develop. A task clustering scheme is devised to preserve the requisite precedence order while honoring timeliness criteria. MSA has been implemented in a POSIX real-time operating environment and its performance profiled under various load conditions. The experimental results show overall performance gains compared to other, memory-unaware UA scheduling algorithms during memory overload. As such, MSA is well-suited for use in a UA governed real-time embedded system. A possible future enhancement to MSA is outlined to produce finer-grain scheduling decisions. Task WCE (Worst Case Execution) times are assumed to be known for UA scheduling. Task WCAR (Worst Case Allocation Requirements) can likewise be determined and incorporated into the scheduling decision. Furthermore, the memory allocation patterns of a task can be profiled as a function of time and the task’s execution divided into storage-based segments, and each segment scheduled individually. The problem of automatic dynamic memory management in a UA environment remains open. This research will next focus on investigating the feasibility, schedulability, utility, and effectiveness of real-time garbage collectors in a UA environment.