Self-adaptation supporting multimedia interaction in
large-scale real-time systems
L. Provensi, F. Eliassen, and R. Vitenberg
Department of Informatics, University of Oslo
Verdione Project
Self-adaptation
Develop a platform that allows people who are
distributed around the globe to interact with each
other as though they were co-located
Ability of a software system to automatically
evaluate its internal state and external
environment and modify its own behavior
accordingly
World Opera
Opera houses spread all over the globe interact to perform
distributed operas
Each house controls a set of subsystems and services offered
to the audience
Sharing of real time media content in a highly heterogeneous
environment
Some examples of self-adaptive behavior for World Opera
Dynamic adjustment of media
quality according to network
condition / device resources
Automatic discovery and configuration
of services for audience using mobile
devices
Management of media streams and
exchange of media flows for other kinds of
lightweight virtual representation
Subtitles
Problems with current frameworks for selfadaptation
Scores
Info
Requirements
Novel mechanisms for self-adaptation that are:
• Flexible – can be customized according to the
needs, of different stakeholders and executing
environment;
• Dynamically tunable – Performance and
resource consumption can be controlled at runtime
• Scalable – Decentralized management of
adaptation services.
• Do not consider the adaptation requirements of all
stakeholders;
• Scalability of a centralized manager in a dynamic
and ultra-large adaptation domain;
• Integration of technologies in a heterogeneous
environment;
• Do not allow reasoning about the adaptation
process.
Solution Outline
Based on two new features that can be integrated into the self-adaptation mechanism:
1 Feedback Control Loop (FCL) design model
The mechanism is designed as a set of interacting
control functions (monitoring, analyzing, planning,
effecting) that can be independently implemented
and deployed.
2 Quality of Adaptation
Qualitative properties of the control functions are
exposed at runtime (response time, memory
usage, etc.) and used to dynamically evaluate the
efficiency of the adaptation mechanism.
MW2
MW5
Adaptation
Services
Adaptation
Services
Analyze
Middleware1
Adaptation Manager
Analyze
Plan
Knowledge
Base
Execute
A1
M
R1
M
Subscribe
Adaptation
Services
D1
Monitor
D2
MW1
Publish
D6
C1
Knowledge
Base Broker
MW4
Publish
Adaptation
Services
M
MW3
Publish
E
D4
Adaptation
Services
D3
Centralized adaptation manager
running on one middleware instance
and executing all control functions
D5
Publish
E
S1
Plan
M
E
P
Decentralized adaptation services running on several middleware
instances and interacting remotely (e.g. using Publish/Subscribe )