http:^^www.cs.utexas.edu^users^ejp^research.html

This data set contains WWW-pages collected from computer science departments of various universities

MIME-Version: 1.0
Server: CERN/3.0
Date: Monday, 06-Jan-97 21:37:41 GMT
Content-Type: text/html
Content-Length: 6549
Last-Modified: Wednesday, 28-Feb-96 00:40:59 GMT

<TITLE> Research Summary for Ed Posnak</TITLE>
<H1> Supporting Adaptive Multimedia Applications in Internetworks </H1>

Recent advances in computing and communication technologies have made
it economically viable to design and implement distributed multimedia
information management systems that promise to enhance users' ability
to access a rich variety of audio, video, and textual information over
globally inter-connected networks. These advances have also resulted
in a networked infrastructure where inter-operating components
frequently differ in capacity by orders of magnitude. Consequently,
multimedia information management systems of the future will be
required to provide services to clients in highly heterogeneous
environments. Traditionally, since functional interoperability has
been the dominant issue, heterogeneity has been sufficiently managed
using black box abstractions (e.g. IP, POSIX).  However, since
differences in performance cannot be hidden by such abstractions, they
are inadequate for meeting the real-time requirements of
resource-intensive multimedia applications.  To address this
performance heterogeneity, there is a trend toward designing
configurable applications that can adapt to the current environment as
well as to the changes in resource availability over time. Mechanisms
for enabling the development of such adaptive multimedia applications
constitute the focus of my doctoral research. <P>

The resource requirements of multimedia applications can vary
significantly depending on presentation processing mechanisms such as
compression and image processing.  Consequently, we have begun
investigating adaptive mechanisms for controlling the cost/performance
tradeoffs of presentation processing.  Specifically, we have developed
the Presentation Processing Engine (PPE), which provides applications
with a compression-independent means for accessing, manipulating, and
changing the quality of media objects.  The PPE is composed of modules
that implement primitive compression components (e.g. Huffman coding,
Discrete Cosine Transform, etc.) and media processing operations
(e.g. scale, clip, etc.).  By allowing basic codec building blocks to
be dynamically configured, the PPE can support different codecs,
switch between resolution levels, and provide flexible control over
QoS parameters such as frame rate, spatial resolution, and Signal to
Noise Ratio (SNR).  The PPE implementation is bound to a scalable
codec at run time based on the media object's compression format and
the application's QoS requirements (expressed in terms of frame rate,
resolution, and SNR).  The implementation can later be dynamically
reconfigured to accommodate dynamic changes in resource availability
as well as the QoS requirements of applications.  The fine-grained
modular architecture allows modules from existing codec
implementations to be suitably extended or reused to implement new
codecs, thereby simplifying software development.  Moreover, support
for media processing operations can be added by plugging in modules
that implement these operations to any codec's internal
implementation.  An advantage of fine-grained configurability is that
it enables such operations to be performed on semi-compressed, as
opposed to uncompressed data, which often yields a significant
performance gain. <P>

Experience developing the PPE toolkit, and using it to implement a
number of codecs has provided some insights into how adaptive
multimedia applications should be built.  Whereas object-oriented
design techniques are powerful tools for building dynamically
configurable implementations, they carry an associated efficiency cost
due to the procedure call overhead of dynamically dispatched method
invocations.  The development of modular, configurable, and efficient
presentation processing mechanisms requires a carefully engineered
balance between the use of static and dynamic composition of modules.
Static composition, implemented in the PPE using parameterized types
and inline methods, minimizes the efficiency cost while maintaining a
reusable and extensible architecture.  For instance, in image and
video compression, the operations to filter a bit stream, get some
number of bits, and decode Huffman symbols should be statically
composed because they are invoked with high frequency, but have
minimal use for reconfigurability.  On the other hand, modules that
perform dithering and the inverse discrete cosine transform are good
candidates for dynamic configurability because they have multiple
implementations with different cost/performance tradeoffs.  By
effectively balancing static and dynamic binding, the PPE
implementations of JPEG and MPEG decoders have achieved performance
that is within 5 percent of the fastest, public domain, monolithic
implementations. <P>

The power of such a dynamically configurable presentation processing
environment depends on the ability to express the configurations as
well as the cost/performance tradeoffs.  We propose to develop an
abstract language for specifying protocol configurations via algebraic
expressions, in which the operators represent modules and the operands
represent data.  These representations will be used by an automatic
configurator to compute appropriate configurations when triggered by a
change in user QoS requirements or by notifications of changes in
resource availability.  The language can also be used to specify
protocol configurations that can be downloaded to the client site
along with the multimedia data.  A key challenge will be to make the
language flexible enough to allow the configuration to be altered to
support media processing operations, while maintaining consistency with
the encode side specification. <P>

The presentation processing environment is an integral component of
the end-to-end systems architecture for distributed multimedia
applications being designed at the Distributed Multimedia Computing
Laboratory at the University of Texas at Austin. Specifically, we
propose to extend our configurable programming environment to support
the QoS-aware transport protocols, and thereby obtain a completely
integrated, configurable protocol stack.  This stack will support
efficient streaming of data from the application to the network
adapter and vice versa, as well as methods for accessing multimedia
objects from our multiresolution file server.  We expect that the
results of this work will substantially advance the state of the art
in building adaptive distributed multimedia applications. <P>