An affinity-driven clustering approach for service discovery and composition for pervasive computing J. Gaber and M.Bakhouya Laboratoire SeT Université de Technologie de Belfort-Montbéliard (UTBM) Belfort, France

2 OUTLINE Context and Objectives Related work Self-Organization Approach to the Design of Emergent Pervasive Services Simulation results Conclusion and future work

3 CONTEXT (1/2) Ubiquitous computing (UC) and Pervasive computing (PC), what’s the difference ? in UC, the objective is to provide users the ability to access services and resources all the time and irrespective to their location. in PC, the main objective is to provide spontaneous services created on the fly by mobiles that interact by ad hoc connections.

4 CONTEXT (2/2) Two new paradigms have been proposed as alternatives to the traditional Client/Server paradigm (CSP) in [GAB00], [GAB06] the Adaptive Servers/Client Paradigm (SCP). the Spontaneous Service Emergence Paradigm (SEP).

5 OBJECTIVES A Self-Organization Approach for service discovery and composition for pervasive applications SDS : Service discovery is the process of locating available nearby services. SCS : Service composition process concentrates in combining different available services discovered by a SDS.

6 RELATED WORK (1/5) Service discovery systems Structured systems Unstructured systems Flooding Random walk Distributed hash tables Indexation Centralized systems Decentralized systems PushPull Parallel random walk Agent cloning

7 RELATED WORK (2/5) Service discovery systems Structured systems Unstructured systems Flooding Random walk Distributed hash tables Indexation Centralized systems Decentralized systems PushPull Parallel random walk Agent cloning Brokers that maintain a repository of published services Hierarchical architecture consisting of multiple repositories that synchronize periodically Cannot meet the requirements of both scalability and adaptability simultaneously The risk of bottlenecks and the difficulty of repositories updating

8 RELATED WORK (3/5) Service discovery systems Structured systems Unstructured systems Flooding Random walk Distributed hash tables Indexation Centralized systems Decentralized systems PushPull Parallel random walk Agent cloning Permits to implement a direct search algorithm to efficiently locate services. Global Overlay network between nodes are generally hard to maintain.

9 RELATED WORK (4/5) Service discovery systems Structured systems Unstructured systems Flooding Random walk Distributed hash tables Indexation Centralized systems Decentralized systems PushPull Parallel random walk Agent cloning Allow nodes to enter and leave the systems without overheads It is not possible to guarantee the success or failure of a query with a constant TTL The mechanism of dynamic TTL or expanding ring is proposed to overcome this problem Generate large loads on the network

10 RELATED WORK (5/5) Service discovery systems Structured systems Unstructured systems Flooding Random walk Distributed hash tables Indexation Centralized systems Decentralized systems PushPull Parallel random walk Agent cloning It is difficult to determine a priori the number of parallel Random walks Agent cloning approach can overcome this problem but need a regulation algorithm

11 SELF-ORGANIZATION APPROACH Service discovery systems Structured systems Unstructured systems Flooding Random walk Distributed hash tables Indexation Centralized systems Decentralized systems PushPull Parallel random walk Agent cloning Self-organization systems Affinity networks

12 SELF-ORGANIZATION APPROACH Objectives: Scalability nodes can establish relationships between them based on their affinity Adaptability affinity relationships between nodes are dynamic; the affinity values can be adjusted at run-time to cope with changes in the environment

13 AFFINITY NETWORKS To build affinity networks, nodes establish affinity relationships between them based on their provided services. Affinity corresponds to the adequacy which two services to bind Adequacy could be implemented based on keywords or objects in common describing a capabilities provided by services. To determine this affinity, services can be expressively described by a language description in order to obtain effective matches between their capabilities (e.g., WSDL).

14 Building and leaving affinity networks let consider D(S i ) a description of the service offered by an Sagent that want to create an affinity relationship with a nearby Sagents. Let us consider also MATSH(D(S i ),D(S j )) a function that return an affinity measure m ij which indicates if the service description of S i matches with the service description of the agent S j. m ij can be calculated as the ratio of keywords that are in common between S i and S i. If m ij is above a certain threshold, agent S i creates an affinity relationship with the agent and S i creates an affinity relationship with S i. An affinity relationship between S i and S i is considered valid if, otherwise, it is discarded and could be removed from the affinity relationship set of S i.

15 AFFINITY ADJUSTMENTS The affinity between two agents is adjusted or reinforced based on two level of satisfaction. local satisfaction: described by services offered by neighboring agents and resources needed to run services (i.e. computing context) global satisfaction: described by the user satisfaction (i.e. user context)

16 SIMULATION RESULTS Simulation using NS2 A network of 100 nodes is generated randomly. Each node provides one service of ten kinds of elementary services that is described by a single of keyword. Each node has no knowledge of services provided by other nodes and the service discovery and composition performs poorly At the beginning of the simulation, there are no relationships, and service discovery and composition performs poorly. As more simulator time elapses, nodes create many affinity relationships with adjustment learning that improve the overall performance

17 CONCLUSION AND FUTURE WORK Decentralized approach for service discovery and composition for pervasive environment is presented. In this approach, the mechanism of establishing affinity relationships is very simple. Other mechanisms can be introduced to increase the rate at which the nodes acquire the relationships that meet the desired and required services. Future work will address the integration of context-awareness parameters in the equations described above together with additional simulations with ns2.