A Dynamic Middleware Control Architecture for Distributed Multimedia
Applications

Baochun Li and Klara Nahrstedt
Department of Computer Science
University of Illinois at Urbana-Champaign
b-li, klara@cs.uiuc.edu


Current state-of-the-art distributed multimedia applications typically
require the underlying environment to provide a desired level of Quality of
Service (QoS).  However, in heterogeneous end-to-end environments,
QoS-aware and QoS-unaware system components may coexist along end-to-end
paths, in which case application demands may not be satisfied continuously,
and QoS violations may occur.  For example, in an end system without
real-time prioritized scheduling and reservation mechanisms for CPU
resources, CPU-intensive applications may not be able to receive constant
QoS with respect to their timeliness requirements in application execution.

The applications, residing in the above described environment, must be
adaptive to the variations in their end-to-end execution. To allow room for
adaptation, we consider the category of applications that have flexible
demands within a specific QoS range.  For these applications, we propose a
Middleware Control Architecture to support the adaptation decisions on
when, how and to what extent these applications may adapt to the
environment.  This architecture serves as a centralized coordinator to
control the adaptation behavior of all concurrent application tasks, so
that if viewed globally, these applications do not adapt in a conflicting
or unfair way.

In order to balance between a centralized controller to enforce fairness
and optimality in adaptation behavior and the diversely different QoS
requirements from different applications, the Middleware Control
Architecture is designed to be consisted of two separate middleware
components, the Adaptor and the Configurator.  Based on control algorithms
derived from the control theory, the Adaptor makes control decisions with
global awareness of application requirements and resource availability of
the entire system [1].  Based on Fuzzy Logic and Fuzzy Control Theory, the
Configurator translates the control decisions made by the Adaptor into
actual parameter-tuning or reconfiguring adaptation actions executed by the
application [2].  Each Adaptor in an end system corresponds to a single
type of resource, such as CPU or network bandwidth, and consisted of an
Adaptation Task, which executes the control algorithm, and an Observation
Task, which monitors the system environment.  Each Configurator corresponds
to a single target application task, and consists of the fuzzifier,
defuzzifier and the fuzzy inference engine that fueled by the
application-specific rule base and membership functions used to define
linguistic values.

On the implementation front, we have implemented a distributed visual
tracking application on Windows NT 4.0 as our testbed, based on the XVision
project [3].  Since the Middleware Control Architecture is designed to
control diversely different applications, we used service enabling
platforms such as CORBA as a medium for all interactions among middleware
components and applications.  Our experiences show that only minor and
trivial modifications are needed in the applications in order to control
their adaptive behavior via our architecture.  We have implemented the
Middleware Control Architecture with Visual C++ 6.0 and Java 2 SDK. We have
identified several adaptation and reconfiguration choices in the tracking
application, such as image size, the number of concurrent trackers,
activation of compression algorithms and the size of actively tracked
region.  Preliminary results have shown that under the control of the
Middleware Control Architecture, the tracking application is able to
maintain a stable tracking precision to track interested objects via
appropriately timed adaptations. We conclude that our approach was valid
and effective in controlling the adaptive actions in applications with
flexible QoS demands. 


References

[1] B. Li and K. Nahrstedt. A Control Theoretical Model for Quality of
Service Adaptations. In Proceedings of Sixth International Workshop on
Quality of Service, May 1998.

[2] B. Li and K. Nahrstedt. Dynamic Reconfiguration for Complex Multimedia
Applications. To appear in Proceedings of IEEE International Conference on
Multimedia Computing and Systems 99, June 1999.

[3] G. Hager and K. Toyama. The XVision System: A General-Purpose Substrate
for Portable Real-Time Vision Applications. Computer Vision and Image
Understanding, 1997.