|
We view autonomic computing as an umbrella term covering
the whole range of technologies, algorithms, techniques
enabling computing systems to regulate themselves much in
the same way our autonomic nervous system regulates and
protects our bodies. We consider the respective computing
model as autonomic computing. We acknowledge autonomic
computing systems as having the following features:
- They have detailed knowledge of their components, current status,
ultimate capacity, and all connections to other systems. They are also aware
of the extent of their "owned" resources, which can be borrowed, lent,
shared or isolated.
- They automatically configure and reconfigure themselves under varying
conditions.
- They are constantly looking ways to optimize their operation, through
monitoring their constituent parts and fine-tuning workflows to better achieve
predetermined system goals.
- They are able to recover from routine and extraordinary events that might
cause some of their parts to malfunction. Moreover, they detect, identify and
protect themselves from attacks jeopardizing their security or integrity.
- They are able to manage themselves in an heterogeneous world, leveraging
open standards. This implies that autonomic computing systems cannot consist of
proprietary solutions.
- They hide their complexity from end users, leveraging the resources to
achieve business or personal goals, wihout involving the user in any
implementation details
Achieving the vision of autonomic computing requires multidisciplinary
research in a broad number of areas. Our group is particulalry interested in
autonomic computing systems that fade into background while providing enhanced
functionality, connectivity, resilience and services to end users. To this end,
we address the research areas outlined in the following paragraphs. The main
research areas of the Autonomic Computing Group are:
Multimodal Technologies and Perceptual Interfaces
Research on Multimodal technologies includes Advanced Digital Signal
Processing techniques for Automatic Speech Recognition using
Microphone Arrays, Face Detection and Face Identification,
Content and Video Classification and processing based on
audio and visual cues. The different modalities may be
supported on a single device or on separate devices working
in tandem. Special Interest is placed in embedding the above
technologies on "thin client" devices having limited power,
computational and presentation capabilities. The target end
result will be non-intrusive multimodal interfaces
transforming meeting rooms, offices and living environments
into smart environments that can interact with the users and
facilitate them in accessing and organizing audiovisual
information in a natural way.
In the area of Multimodal Technologies the Autonomic
Group is already participating in the
CHIL EC funded project. AIT is also pursuing
collaborations with leading companies and research
institutions in the area including IBM and CMU.
Since 2003 and for the purpose of various project and
research activities the following components were researched
and developed
- FACE DETECTOR: a Perceptual Component that can
detect the exact position of the human face inside a
picture.
- FACE RECOGNIZER: a Perceptual Component that takes
the information of the face detector and analyzes it in
order to end up recognizing the person on the image.
- TRACKER: a Perceptual component that tracks moving
targets in the camera’s field of view. By combining
information from more than one camera we can also
achieve a 3D-TRACKER that can give exact 3d coordinates
of a moving person.
- ASR(AUDIO SOURCE LOCALISATION): a Perceptual
component that uses the information of two microphones
and locates the source of the speaker. Combining more
pairs of microphones in different locations we can also
locate the exact 3D-Location of the speaker.
- BSS(BLIND SOURCE SEPARATION): a Perceptual Component
that can separate N speakers speaking simultaneously
using the information of N microphones
- SAD(SPEECH ACTIVITY DETECTION): a Perceptual
Component that can disseminate intervals that contain
speech from those that contain noise.
- SPEAKER IDENTIFICATION: a Perceptual Component that
can detect the speaker in an audio segment.
- SPEAKER DIARISATION: a Perceptual Component that
inside an audio segment can decide at which points we
have the change of a speaker and whether this speaker
has already spoken inside this segment.
Autonomic Middleware for Perceptual Systems
Closely related with the above areas and in order to provide the
infrastructure for the components above to be able to
communicate, exchange data and information, is our research
on advancing the middleware. We strongly believe that it
will evolve from the state of art XML messaging
infrastructures (e.g., Web Services) to more advanced
infrastructures (such as knowledge base) keeping track of
state and providing a host of non-functional features such
as resource management, scalability, fault tolerance. Based
on such components we intend to offer integration of
perceptual interfaces components, ubiquitous mechanisms for
accessing resources, as well as intelligent techniques for
managing resources.
The recent researched and already developed architectures
are:
- CHIL ARCHITECTURE an architectural framework, along
with a set of middleware elements facilitating the
integration of perceptual components, sensors, actuators
and context-modeling scripts comprising sophisticated
ubiquitous computing applications in smart spaces. The
architecture puts special emphasis in the integration of
perceptual components contributed by a variety of
technology providers, which has not been adequately
addressed in legacy architectures
- BODY TRACKER API is a middleware API that consists
of structured XML schemas, defining input and output
parameters for each component.
Grid Computing
The evolution of the Grid to a Global Computing Infrastructure requires
multidisciplinarity touching a broad spectrum of areas. We
believe that functionality is required in multiple layers,
each one giving rise to a particular Grid research area
Specifically, the Grid implementation involves:
- Selecting, designing and engineering the infrastructure resources,
including networking, servers, storage and operating systems that have the
qualities to support the Grid.
- Establishing a distributed system across the multiple heterogeneous
resources, and accordingly support functions such as scheduling, global resource
discovery and consistent authentication across multiple administrative domains.
- Providing a distributed programming model, to be used for producing Grid
applications.
- Producing applications that can exploit the Grid infrastructure, while also
being able to interact with each other based the distributed programming model.
- Resolving administrative and management issues relating to users' accounts,
security policies, business models, target operational processes.
The group will investigates the full suite of the issues outlined above,
starting from infrastructural issues and ending up with business models and
policies. Currently the research emphasis is on models predicting resources
utilization (e.g., host load, physical memory) towards devising and developing
scheduling algorithms for Grid infrastructures comprising highly heterogeneours
donors. As a next step we aim at combining these algorithms with resource
allocation schemes ensuring the Virtual Organizations contributing resources to
a Grid, enjoy a fair share of resource for own use. Our implementation work is
based on the convergence of Grid Services with Web Services (i.e. GT3.x, WSRF).
AIT Autonomic Group is closely collaborating with GRNET towards monitoring and
participating in the evolution of Grid infrastructures and applications within
the EU. Further, we are collaborating with SUN Microsystems and IBM on
applications and technology integration issues, under the broader umbrella of
High Performance Technical Computing.
Recent achievement is the EMPEROR that provides a
framework for implementing scheduling algorithms based on
performance criteria. In implementing a particular
instantiation of this framework, we have devised models for
predicting host load and memory resources, and accordingly
for estimating the running time of a task.
Broadband Network Control
The members of the Group have a rich background in devising mechanisms for
network control (admission control, traffic engineering,
routing, resource management), through producing algorithmic
schemes and integrating them to traffic control frameworks
mainly targeting network QoS. Despite the bandwidth offered
by state of the art optical network, wireline networks are
still in need of such mechanisms, in order to:
- Provide seamless inter-domain seamless QoS support, for consistent
end-to-end services
- Control user access to services (since accounting, authorization and
authentication is needed, without agreement on inter-domain policy management
architectures)
- Enable Service monitoring (point-to-point and end-to-end) and provide
feedback to applications of current service level (end-to-end QoS), application
adaptation
Therefore, the group deals with the following research topics relating to
Broadband Network Control, targeting mainly IP based networks (i.e. packet
level):
- Traffic Modelling
- Traffic Control Frameworks
- IP Architecture Evolution - (includes IPv6)
- Analytical and Simulation Models and Performance Analysis
- SLA/SLS Management
- Interworking of IP networks with Optical, Wireless
- End to End QoS for IP Networks
- Adaptable/Customizable QoS
|
