Project topics and materials
Get Complete Project Material File(s) Now! »
Architectures for autonomic systems
The reference architecture for autonomic managers is that of a feedback control loop mapping observations on managed resources and environment to control orders, according to a high-level objective. The control loop model has been refined in the seminal Vision of Autonomic Computing with the MAPE-K architecture [KC03], presented on Fig. 2.1. MAPE-K divides the control loop into 5 logical components:
⋄ Monitoring: gather information through available sensors on the managed resource and possibly in its environment, and produce symptoms.
⋄ Analysis: read available symptoms and issue change request so as to follow the high-level management goals issues by the user.
⋄ Planning: handle incoming change request by producing change plans to be executed on the managed resources.
⋄ Execution: execute change plans through effectors on the managed resource.
Taxonomy of autonomic systems
Once established a qualitative definition of autonomic systems, it can be enriched with quantitative “degrees of autonomicity” a system can feature. IBM provided its own metric for autonomic systems, ranking them on 5 consecutive levels along two orthogonal dimensions [IBM06]:
⋄ on one axis, the “autonomic functionality” falls into five consecutive categories:
– manual : the system has no support for autonomic management and is administered by hand.
– instrument & monitor : the system provides administrator with management interfaces, i.e. sensors, effectors and limited monitoring functionalities.
– analysis : the system can provide high-level representation of its situation, and possibly reconfiguration advice.
– closed loop : the system can apply itself some reconfigurations automatically, but still under the administrator’s control.
– closed loop with business priorities : the system does not depend on administrators any more, and communicates with them through high-level business objectives.
Lack of generic solution to complexity
In this section, we reviewed how state-of-the-art autonomic systems tackle the concerns of dynamism, openness, heterogeneity, scalability and multiple objectives. To our knowledge, each of these complexity factors has been tackled individually by significant contributions in the domain. However, these contributions have not been generalised and made reusable for different use cases, and none of them addresses all the identified complexity factors. In addition, as previous section shown, little generic support is available to integrate existing partial solutions into a unified approach to complex autonomic systems. In this thesis, we propose a basis for addressing these concerns via a generic support for the design of complex autonomic systems. We illustrate our generic contribution on the particular case of smart electricity grids, that next chapter presents as a relevant example of complex system.
Table of contents :
1 Thesis overview
1.1 Context
1.2 Challenges
1.3 Contributions
1.4 Thesis outline
2 Background in Autonomic Computing
2.1 Definitions
2.1.1 Self-* properties
2.1.2 Alternative definitions
2.2 Autonomic system implementation
2.2.1 Architectures for autonomic systems
2.2.2 Taxonomy of autonomic systems
2.3 Examples of autonomic frameworks
2.3.1 Rainbow
2.3.2 Autonomia & Automate
2.3.3 SMC & Ponder2
2.3.4 MACODO
2.3.5 Ceylon
2.4 Limited support for complexity in existing autonomic systems
2.4.1 Complexity factors in autonomic systems
2.4.2 Complexity factors addressed by state-of-the-art autonomic systems
2.4.3 Lack of generic solution to complexity
3 An introduction to Smart Grids
3.1 Power management in smart grids
3.1.1 A simplified grid model
3.1.2 Power management in the grid
3.2 Smart micro-grid scenarios
3.2.1 Heating system scenario
3.2.2 House scenario
3.2.3 District scenario
3.3 Requirements for the grid’s control infrastructure
3.3.1 Complexity Requirements
3.3.2 Integration Requirements
4 Thesis position
4.1 Challenges addressed
4.1.1 Need for generic architectures and conceptual frameworks
4.1.2 Need for an integration-oriented approach
4.1.3 Challenges in the case of smart micro-grids
4.2 Contributions
5 A generic formalisation of goals for multi-objective autonomic systems
5.1 Goal formalisation
5.1.1 Viability constraints
5.1.2 Scopes
5.1.3 Two sample management objectives
5.1.4 Usage and utility of goal formalisation
5.2 Goal fragmentation and composition
5.2.1 Goal splitting
5.2.2 Goal translation
5.2.3 Goal conflicts and resolution requirements
5.3 Application: holonic architecture for grid control
5.3.1 Goal specification for power management
5.3.2 Goal specifications for electric appliances
5.3.3 Integration requirements
6 A generic architecture for adaptable, multi-objective autonomic managers
6.1 Taxonomy of autonomic management components
6.1.1 Fragmentation and integration for modular autonomic managers
6.1.2 Control components
6.1.3 Integration components
6.2 Task integration patterns
6.2.1 Monolith pattern
6.2.2 Aggregator pattern
6.2.3 Dealer pattern
6.2.4 Arbiter pattern
6.3 Application: the smart heater use case
6.3.1 Single-objective autonomic heater
6.3.2 Multi-monitoring extension
6.3.3 Multi-objective extension
6.3.4 Extending the approach to more complex management cases
7 Generic integration patterns for open organisations of autonomic managers
7.1 Organisations overview
7.1.1 Abstract organisation architecture
7.1.2 Organisations in the literature
7.2 Organisational patterns
7.2.1 Hierarchy pattern
7.2.2 Stigmergy pattern
7.2.3 Collaboration pattern
7.2.4 Organisation evaluation and adaptation
7.3 Application: power management organisations
7.3.1 Generic organisation specification
7.3.2 Conflict resolution via scheduling in power management organisations
7.3.3 Integration of organisations at different holonic levels
7.3.4 Possible organisation variants
8 Proof-of-concept implementation and experiments
8.1 Simulation features
8.1.1 Temperature
8.1.2 Electric network
8.1.3 Appliances
8.2 Simulator Implementation
8.2.1 Main software technologies
8.2.2 Smarduino House
8.3 Implementation of Managers and Power Management Organisations
8.4 Experimental Scenarios and Results
8.4.1 House scenarios and results
8.4.2 District scenarios and results
9 Conclusions & Future Work
9.1 Summary of Thesis Contributions
9.2 Limitations and Future Work
9.2.1 Towards a Generic Framework for Autonomic Systems
9.2.2 Support for Time-related Concerns in Autonomic Managers and Organisations
9.2.3 Meta-Management
9.2.4 Autonomic Societies
9.2.5 Ergonomics of Autonomic Systems
9.2.6 Scheduling Algorithms for Power Management
9.2.7 MisTiGriD 2.0
9.2.8 Autonomic Management at Every Grid Level
9.2.9 Holonic Patterns
9.2.10 Application to Other Domains
Bibliography
