A Machine Learning Middleware Service for OnDemand Grid Services Engineering and Support

1
A Machine Learning Middleware Service for
On-Demand Grid Services Engineering and Support

• Prof. A. Taleb-Bendiab
• School of Computing
• Liverpool John Moores University
• email a.talebbendiab_at_livjm.ac.uk
• http//www.cms.livjm.ac.uk/taleb
• http//www.cms.livjm.ac.uk/Self-X

2
Scope

• Situated Autonomic Computing
• Problem Definition - Challenges
• Design including coordination and control
• model-based vs emergence
• Specification of control models
• Design via experimentation and machine learning
• Example on-demand reservation of application
  services
• User Classification scenario
• Episodic resource requirements
• SOM Classification for Connected Home Machine
• Implementation
• Case-study

3
Situated AC Scenario E-Fire Services
4
Challenges -- Global Computing

• Global Enterprise Systems
• High-assurance systems development and life-time
  management
• Complexity and scale is rapidly increasing
• Bio-inspired Models -- Autonomy
• devolving software management, maintenance to the
  software itself
• Self-managing, self-tuning, self-protecting, ...
• Need continuous measurement, introspection to
  support
• Observed and/or supervised adaptation for
• Safe, predictable,
• Coordinated, traceable, etc.

5
So far !

• Current research
• Instrumentation middleware services for
• improved usability and reliability for instance
  for
• grid-based applications, and ubiquitous computing
• Monitor, control and manage grid users
  applications.
• Context-awareness and QoS-Aware systems
• Event-based systems
• Sensor networks, Etc.
• Further research is required
• Management, assurance and fidelity of awareness
  layer is a major concerns
• Sensors and actuators (effectors) support web
  services and grid computing
• Current models looking at small scale systems

6
Design Approach Informed by Machine Learning

• Frameworks and Models
• Programming, interaction and/or control models.
• Two experiments were conducted
• User Classification and on-demand service
  reservation
• Autonomic software restore service

7
Experiment 1 User Classification

• The scenario
• Mining service usage models per class of users
  for preemptive service reservation and on-demand
  services
• Method
• Developed an Simulation tool for Intelligent
  Connected Home, which generate services
• Self-Organising Maps (SOM) applied extract
  feature or usage model
• Design and Implementation
• To follow

8
Design and Implementation

• Data generated tool is developed to produce
  training and test data for this application.
• An OGSA and web service compliant SOM middleware
  service was developed
• For rapid prototyping a Matlab library for SOM is
  used for classification

9
SOM Classification Results For Connected Home
Machine Devices

• Lights and PlayStationII correlates
• Video and Coffee Machine correlates
• Video CD and Fans correlates
• Vacuum cleaner and Washing machine correlates

10
ML Middleware Services
11
So What?

• Exploiting ML
• anticipate and organize the consumers requests
  in advanced.
• Job schedule is responsible for managing the
  invocations of the services.
• Just-in-time services invocation and usage
• Etc.
• In addition to the presented ML middleware
  service with automated inclusion and use of usage
  model for user and service classification
• Further support is required including
• Specification and modelling of mined models and
  their enactment for instance
• Control and/or actuation models
• Neptune Meta-Language and Integrated development
  environment will be used for this.

12
Neptune Meta-Language 1
13
Neptune Meta-Language 2
14
Neptune Meta-Language 3
15
Conclusions Further Work

• Prototypes developed using .Net and support Web
  Services Standards
• Tested in a number of case studies
• Intelligent Connected Homes
• E-Health
• With PlanetLab environment
• Further work
• Integration of this work with the Neptune
  Language to support
• norm-governed web services and architectures.
• Situated Autonomic middleware
• Integration machine learning services to support
  danger/novelty detection
• Further evaluation of the framework

16
THANK YOU