The Hong Kong University of Science and Technology
Department of Computer Science


PhD Thesis Defence


"Power-Aware Scheduling in Computing and Communications with QoS
Requirements"

By

Mr. Fan Zhang


Abstract

Driven by the demand for higher computational power and communication
capabilities, the power consumption of electronic computing devices has
increased drastically in the past decade. This calls for innovative and
more energy efficient system designs not only for battery-powered portable
devices but also for computer servers and server clusters that face the
overheating problem. The concept of power-awareness has been incorporated
into today's hardware and software development as an important design
criteria.

This thesis studies the use of software strategies to improve energy
efficiency in computing and wireless communications. In particular, we
investigate the impact of scheduling on power management and propose new
strategies for processor scheduling and wireless data
transmission/reception to improve energy efficiency without violating the
QoS requirements of the applications.

We first address the issue of power-aware voltage scaling for real-time
periodic tasks. The concept of "blocking-awareness" is introduced in the
context of scheduling tasks with non-preemptible sections. The proposed
blocking-aware algorithms dynamically adjust the processor speed based on
run-time blocking occurrences instead of worst-case blocking expectation,
therefore they can greatly reduce processor energy consumption over static
speed schemes.

The workload of desktop PCs and server systems is usually aperiodic and an
average delay guarantee is desirable. We design stationary voltage scaling
policies that adjust the processor speed according to the number of tasks
in the system. Markov models are employed to compute the statistically
energy-optimal policy. This approach is then extended to multiprocessor
systems and heterogeneous server clusters, where some processors are
selectively turned off to further reduce energy dissipation.

The wireless transceiver is another major energy consumer in a networked
computer system. For wireless networks with limited energy and a designed
lifetime, it is important to maximize the utility of available energy
instead of minimizing energy usage. We present an iterative algorithm to
compute the optimal transmission power setting to multiple receivers. The
transmission schedule produced fully utilizes the available energy and
achieves maximum data throughput in each operating cycle. When data
transmission is associated with values, the value maximization problem is
formulated as a convex optimization problem, and is solved efficiently
using a dynamic programming approach.

Finally, we analyze the energy consumption of data reception in a wireless
LAN environment. We point out that existing standards are not energy
efficient as they may cause mobile clients to wake up frequently and wait
for an extended period of time before actual data reception. Considering
streaming media applications over the Internet, we propose schemes that
combine power-aware scheduling and traffic shaping at the local proxy
server. The proposed schemes minimize the waiting time of the mobile
clients and effectively reduce their total energy consumption.
Furthermore, we extend our algorithms to take the residue battery
capacities of the clients into consideration so that clients at low
battery level could spend less energy and enjoy a prolonged battery life.

It is our wish that these analyses and algorithms will benefit the design
of energy-efficient computing and wireless communication software systems.


Date:			Wednesday, 13 April 2005

Time:			10:30a.m.-12:30p.m.

Venue:			Room 2404
			Lifts 17-18

Chairman:		Prof. Tai-Kai Ng (PHYS)

Committee Members:	Prof. Samuel Chanson (Supervisor)
			Prof. Lionel Ni (Supervisor)
         		Prof. Mounir Hamdi
			Prof. Jogesh Muppala
                        Prof. Chi Ying Tsui (ELEC)
                        Prof. David K.Y. Yau (Comp. Sci., Purdue Univ.)


**** ALL are Welcome ****