Power-Aware High-Performance Computing: The Dawn of the Green Computing Era 

Ishfaq Ahmad, University of Texas at Arlington 

Abstract 

Energy is one of the most valuable and scarce resources, a significant portion
of which is now being consumed to power up computers and their accessories.
Energy and cooling are the two of the biggest issues facing IT organization
today and a growing number of companies need ways to curb these costs while
enabling future expansions. A side effect is an escalating threat to the
environment. However, energy saving usually comes at the expense of
performance. In this talk, we give an overview of our research activities
related to various aspect of energy and performance optimization, as well as
their trade-offs. We address several research issues in power-aware computing
at various levels, such as system, software, and algorithm. 

The extensive computational requirements of various applications on grids make
energy a critical resource in their execution. We propose scheduling algorithms
and tools that can be used by grid managers or service providers to save energy
while ensuring performance. The design of such algorithms and resource
management schemes poses new research challenges because they must deal with
the dynamic nature of the grid and task-to-machine matching requirements, while
considering complex service policy issues as well as both short term and
long-term energy goals. We propose game theory based scheduling algorithms that
optimize both energy and quality of service. Our game theoretic techniques not
only yield efficient solutions for this problem with low overhead, but are also
flexible to self-tune for accommodating various problem scenarios and
management policies. 

We also address the energy problem from an application point of view. In mobile
environments, multimedia applications running on high-performance embedded
systems are faced with the problem of limited battery power. For instance,
video compression, due to its intensive computational requirements, can quickly
deplete a battery. The theoretical basis of current video compression
technologies is the quintessential R-D (rate-distortion) model that epitomizes
the non-linear relationship between distortion and target bit rate, but does
not address the energy problem. We propose a new model that allows a video
encoder to simultaneously address the energy consumption, visual quality and
bitrates. The model enables algorithmic complexity control to achieve energy
and performance goals in video coding. A software-based architecture is also
designed that allows the proposed techniques to be used in conjunction with
MPEG and H.26X video coding standards. 

Biography 

Ishfaq Ahmad received a B.Sc. degree in Electrical Engineering from the
University of Engineering and Technology, Pakistan, in 1985, and an MS degree
in Computer Engineering and a PhD degree in Computer Science from Syracuse
University, New York, U.S.A., in 1987 and 1992, respectively. He is currently a
professor of Computer Science and Engineering at the University of Texas at
Arlington (UTA). At UTA, he leads the Multimedia Laboratory and the Institute
for Research in Security (IRIS). IRIS, an inter-disciplinary research center
spanning several departments, is engaged in research on advanced technologies
for homeland security and law enforcement. Professor Ahmad is known for his
research contributions in parallel and distributed computing, grid computing,
multimedia computing, video compression, and security. His work in these areas
is published in close to 200 technical papers in peer-reviewed journals and
conferences, including three best paper awards at leading conferences and 2007
best paper award for IEEE Transactions on Circuits and Systems for Video
Technology. His current research is funded by the Department of Justice,
National Science Foundation and industry. He is an associate editor of the
Journal of Parallel and Distributed Computing, IEEE Transactions on Circuits
and Systems for Video Technology, IEEE Transactions on Multimedia, IEEE
Distributed Systems Online, and Cluster Computing. In recognition of his
contributions in parallel computing, he became an IEEE Fellow this year.