Thursday, December 5, 2002
Noon - 1:30 pm
Intel Seminar (417 S. Craig Street - 3rd Floor)
EVENTS PAGE: http://www.intel-research.net/pittsburgh/events.htm
Department of Computer Science, Duke University
Durham, NC, USA
Supporting Variable Bit-Rate Streams in a Scalable Continuous Media
Dramatic advances in data storage and networking technology have not been
sufficient to provide inexpensive wide-scale distribution of streaming
video with acceptable quality.
Variable bit-rate encoding is known to generate media streams of considerably
smaller size than constant bit-rate encoding of equivalent quality. Hence,
the capability to support variable bit-rate streams in content distribution
networks has the potential to significantly reduce the cost of media streaming,
due to lower requirements for bandwidth and storage space.
In this talk, we describe the design of Exedra, a scalable continuous
media server architecture, and its prototype implementation. Innovative
resource management policies are examined under a realistic performance
evaluation methodology. System throughput is shown to
increase almost linearly with the number of disks, while offering deterministic
quality of service guarantees. Fault-tolerant operation can be enabled
with minimal additional cost.
High variability in resource requirements can also lead to reduced utilization
of system resources. We formulate the need for balancing the disk bandwidth
against server buffer utilization as a smoothing optimization problem.
A novel optimal algorithm is introduced that can be successfully applied
to streams stored across either homogeneous or heterogeneous disks.
Stergios Anastasiadis received MSc and PhD from the Department of Computer
Science, University of Toronto in 1996 and 2001 respectively, and Dipl.Eng.
from the Department of Computer Engineering and Informatics, University
of Patras, Greece in 1994. Since September 2001 he has held a Visiting
Assistant Professor position at the Department of Computer Science, Duke
University. In his doctoral dissertation, he investigated system design,
resource management, and performance evaluation issues in scalable streaming
media servers. Previously, he also worked on parallel application scheduling
for cluster-based multiprocessors.
Contact Kim Kaan, 412-605-1203,
or visit http://www.intel-research.net.
SDI Home: http://www.pdl.cmu.edu/SDI/