Abstract:

Due to the proliferation of the battery-operated embedded devices,
energy has been recognized as a first-class resource in computing
systems and many software/hardware power management techniques have
been studied. As one of the most popular power management techniques,
dynamic voltage and frequency scaling (DVFS) scales down the supply
voltage and operating frequency of modern processors for non-peak
workload to save energy. However, recent studies show that DVFS has an
alarming negative effect on system reliability due to increased
transient faults (especially for the ones induced by cosmic ray
radiations) at lower supply voltages.

In this talk, I will present our recent work on the reliability aware
power management (RAPM) techniques for real-time embedded
systems. Focusing on tolerating transient faults, the central idea of
RAPM is to schedule necessary additional recovery tasks to preserve
the desired system reliability using the rollback recovery
techniques. After illustrating the basic idea of RAPM for a single
real-time tasks, I will present the RAPM schemes for a set of periodic
real-time tasks. At the end, exploiting the probabilistic execution
time information of real-time tasks, our latest work on optimistic
RAPM schemes will also be discussed.

Short Bio:

Dr. Dakai Zhu is currently an Assistant Professor at the University of
Texas at San Antonio (UTSA). He got his Ph.D. at the University of
Pittsburgh in 2004. He got his M.S. at the Tsinghua University and
B.S. at the Xian Jiaotong University in 1999 and 1996, respectively.
His research focuses on real-time embedded systems and low power
computing. He has co-chaired three workshops and served in the program
committee for many conferences (e.g., RTAS, RTSS and ICESS). He is a
member of IEEE.