The Department of Computer Science, as part of the CS Mentoring Series, presents a talk by Aaron Block '02 Ph.D. Candidate in Computer Science University of North Carolina - Chapel Hill Adaptive Multiprocess Real-Time Systems and Multimedia Applications

Aaron Block

In this talk, I will summarize my dissertation research in which I extend multiprocessor real-time systems so that such systems could change the share of a processor used by a real-time "task" at run- time -- a process called reweighting. The particular focus of my work is reweighting schemes that are deployed in environments in which tasks may frequently request significant share changes. Such environments are commonplace in computationally- intensive multimedia applications. Currently, there are no multiprocessor reweighting algorithms that are capable of enacting task share changes with bounded overhead. In my research, I built upon prior work on uniprocessor and multiprocessor systems to construct reweighting algorithms with minimal overhead for both partitioned systems, in which tasks are permanently assigned to each processor, and global scheduled systems, in which tasks can freely migrate between processors. Furthermore, I examined how feedback-control techniques can be used to determine the task shares at run time. Finally, I tested these algorithms using two multimedia applications, Whisper, a human tracking system, and the Virtual Exposure Camera, a night vision system, both developed at The University of North Carolina at Chapel Hill.