Research On Key Techniques Of Self-adaptative Service Discovery In Pervasive Computing

After the mainframe computing paradigm, desktop computing paradigm anddistributed computing paradigm, the computing paradigm is coming to pervasivecomputing (PvC) paradigm. Pervasive computing seamlessly integrates the real worldand the information space into a space, in which people can obtain the necessaryinformation and services anytime and anywhere. In the enjoyment of servicesprovided by pervasive computing, challenges are brought to the service discovery inpervasive computing.The diversity and dynamics of pervasive computing environment (PvCE)demand the adaptability of service discovery in PvC, and the autonomy of PvCEchallenges the construction of the service discovery protocol (SDP). To run in thePvCE, a service discovery protocol should be able to adapt to devices with differentabilities, networks with different features and users with different intentions, andadapt to the dynamics of them. The autonomy of devices in PvCE often leads toselfishness, which makes node lack of motivations to follow the rules defined bySDPs. Selfish nodes even try to break the rules to maximize their own interests.Therefore, service discovery protocol needs a mechanism to adapt to the autonomy ofnodes.Although the service discovery is a hot issue in PvC research, researchers payfew attention to the adaptability of SDP. They are more interested in the constructionof protocols for specific environments such as mobile ad hoc network, and otherfeatures of SDPs such as semantics, performance, security. Two factors hinder thetraditional service discovery protocol to obtain more adaptability:1) the fixedfunction set and the closed interaction rule set that the roles in SDPs impose on thedevices,2) the dilemma of assuming underlying network features. Existing methodssuch as protocol interoperating are difficult to systematically meet the challenge ofSDP adaptability raised by the diverse, dynamic and autonomic PvCE.To meet the challenges of PvCE to the adaptability of SDPs, this dissertationprovides a quantitive method to analyse and evaluate the adaptability of SDPs, basedon the clearly defined concept of adaptability of SDPs. In the application of thequantitative method, two inspirations to enhance the adaptability of SDPs arise:1)online feedback and adjustment,2) open management and control. With the secondinspiration, we analyse the reasons that hinder the SDPs to acquire more adaptability, and propose an adaptive VGAP framework for service discovery. According to thefirst inspiration, an self-adaptive Market model is constructed for service discovery,by analogy with the market in economics. Appling the Market model to the VGAPframework, we get a self-adaptive service discovery schema VGMAP. The maininnovation of this work include:(1) A quantitative method to analyse and evaluate the adaptability of SDPsBased on the similarity between SDPs and economic systems, this paperproposes a quantitive method, derived from the Cost-Benefit Analysis in economic, toanalyse and evaluate the SDPs’ adaptability, which is clearly defined at first. With theproposed quantitive method, objective evidences can be acquired by calculation andstatistics, conclusions can be drawn with the accurate criterions and measure methods,and the weaknesses of the SDPs can be uncovered by comparing the criterions. Thequantitive results are able to direct us to select and enhance the SDPs in pervasivecomputing. By applying the quantitive method, we analyse and evaluate some simpleSDPs. With the results produced by the quantitive method, we compare theiradaptabilities and overcome their weakness uncovered. In the process of analysis,evaluation and comparison, we get two inspirations on enhancing the adaptabilities ofSDPs:1) online feedback and adjustment,2) open management and control.(2) An adaptive ability-based and policy-driven service discovery frameworkVGAPMost of the existing SDPs provide the service discovery utilities by defining thetwo sets of functions and interaction rules on the devices. Since the sets of functionsand interaction rules defined at protocol design time are fixed and closed, it will bedifficult for them to adapt to the diverse and dynamic PvCE. Some SDPs assume theunderlying network features for some reasons also hinder them too much to adapt tothe PvCE. We proposes the ability-based and policy-driven AP model to overcomethe restraints of the two fixed and closed sets on the SDPs’ adaptability. Moreover, APmodel is extended to the VGAP framework, which enables SDPs to take fulladvantage of network features without affecting the ability of adapting. The secondsuggestion can be applied to the VGAP framework, enhancing its adaptability.(3) An economic market based self-adaptive Market model for service discoveryThe policy selection in VGAP framework is a distributed constraint optimizationproblem (DCOP) with some features, which make it difficult for traditionalalgorithms to solve. Based on the market mechanism in economics, a Market model isbuilt for the nodes in VGAP. In the Market model, each node should solve three local constraint optimization problems, according to the hypothesis of rational economicman. The three constraint optimization problems are: to maximize the profit whenselling resources of the node, to maximize the success expectation when select thecandidate policies to handle events, to maximize the user utility when invest money tothe requests or events. By applying the Market model to VGAP, the policy selectionDCOP is solved and the lack of motivation to obey protocol rules is overcome, we geta self-adaptive service discovery schema VGMAP.(4) The algorithm for local constraint optimization problems in Market modelThe three local constraint optimization problems in the Market model haveunknown and dynamic functions in their composited target functions. Moreover, itwill be expensive to evaluate the value of the target function in the three problem.Using support vector machine derived from the statistical learning theory and probingthe optimal points randomly, an online optimization algorithm is proposed to solvethe three problems satisfying the PvCE.To validate the above works, the VGMAP schema is implemented in a prototypepervasive computing platform. At the same time, a generic discrete event orientedsimulation framework is presented to make it easy to verify the theories in thisdissertation by rapid simulations.