Date: Thursday, February 22, 2001
Time: 3:30 P.M. (Talk)
Place: Withers 402-A, NCSU Historical Campus (click for
courtesy parking request)
Speaker: H. Aydin, Computer Science, University of Pittsburgh
Reward-Based Real-Time Scheduling: A Framework of Tradeoffs for Timeliness, Precision, Power Efficiency and Reliability
Abstract: Reward-Based (RB) Real-Time (RT) Scheduling is a resource management framework allowing to trade precision for timeliness. In this talk, in addition to presenting the optimal and efficient solution to the optimal periodic RBRT Scheduling problem, we describe how the framework can be used to express and exploit tradeoffs on additional requirement dimensions such as power-efficiency and reliability without compromising timeliness.
In RBRT scheduling model, each real-time task is logically decomposed into a mandatory part and an optional part. A reward function is associated with the execution of the optional part. In the first part of the talk, we discuss the solution to the optimal periodic RB scheduling problem which was open prior to our work. The solution and its important implications are based on a main result which states that, for the most common (i.e. linear and concave) reward functions, there exists always a schedule where a given task receives the same amount of optional service time from instance to instance.
Dynamic Voltage Scaling (DVS) is a technique based on reducing the speed (hence the power consumption) of the processor on-the-fly, at the expense of increased latency. The RT-DVS problem is to minimize the energy consumption through the DVS, while still meeting all the deadlines. After describing how the static version problem of the problem can be casted (and solved) in RBRT framework, we show that reclaiming unused computation times and further, speculatively reducing the CPU speed can provide important power savings.
Our work on the reliability dimension stems from the observation that in RBRT models, mandatory parts have still hard deadlines and they must be completed before the deadline even in the presence of faults. We present our Fault-Tolerant Optimality framework where the objective is to produce a schedule which tolerates the transient faults of mandatory parts while maximizing the total reward.
Short Bio: Hakan Aidin received B. Sc. and M. Sc. degrees in Control and Computer Engineering from Istanbul Technical University in 1991 and 1994, respectively. From 1994 to 1996, he was the Chief System and Network Administrator at the Computing & Network Services Center, Istanbul Technical University.In 1996, he joined the Computer Science Department at the University of Pittsburgh as a graduate student, where he is currently a Ph.D. candidate. He is a program committee member of IEEE Real-Time Technology and Applications Symposium (RTAS 2001). His research interests include real-time systems, low-power computing and fault tolerance.
Hosts: C. Healey,
Computer Science, NCSU