Research On Dynamic Updating Ofservice-Oriented Distributed System

Currently in large-scale distributed online service system, reliability, availability, credibility and maintainability of computer systems have gradually become important factors to be considered, especially in a number of key application areas such as: financial data processing systems, medical service systems, remote communication systems, and telecom online billing system. In such distributed systems, real-time decision supporting or event-driven processing is often necessary. Once the system begins running, maintenance operation for 24 hours a day, seven days a week is needed. System's interruption means a high price and even brings life-threatening to users. Therefore, dynamic updating of software is becoming an important research topic in the field of software maintenance technology now. Whether or not to support uninterrupted service is becoming an important indicator for a software system.As service-oriented software design method has advantages in dynamic updating, interactivity and other areas, to build applications based on service has become a trend in software engineering. Therefore, how to support dynamic updating in a service-oriented distributed system is a topic worthy of in-depth study. This thesis will put forwards critical issues to be solved for supporting dynamic update in"24*7"service-oriented distributed system.Because service-oriented software design method gives more support on dynamic and interchangeability, and component-oriented technology pays more attention to reusability of software module, many of the popular software technologies are based on service-oriented component model (SOCM). And such systems based on SOCM introduces the concept of service-oriented into the component-oriented model. Therefore, research of this thesis is carried out based on SOCM.Main research and innovation in this thesis includes the following aspects:1,Analyze deeply services'remote access features, including the release of the remote service, requests of remote service, invocation of remote service, execution and return of remote service. And then make in-depth analysis of issues about determining of the safe updating-time, the transmission of status, rebinding of the updated service, and coordination of multi-services'updating.2,According to transactional characteristics of runtime-services in a distributed system, a group of control strategies for dynamic updating are proposed. These strategies include: selection of slow or fast update strategy according to transaction's long or short feature; a strategy for selection of updating time, which is determined according to whether or not related short transactions has finished(the related short transactions are a single or multiple transactions, which need updated services to participate in. This method needn't forcefully stop the update services'running.), and a carrying-out-updating strategy based on ACID of short transaction, which proposes to use a short transaction model to control the basic update operations, and uses a long transaction model to control persistent data conversion. In all related updating operation step, persistent data conversion will usually take the longest time. And furthermore, a lazy approach to the implementation of persistent data transformations is proposed, and the serialization of conversion transactions and application transactions in the intersection caused by this lazy approach is given to ensure the consistency of the system state before and after updating.3,A hierarchy coordination model CoDUM-TrSOS for updating based on transaction is given to support the dynamic updating of system. Model design mainly includes three parts: the design of single component containers (including the atomic service-component update-container and composite service-component update- container), a multi-service-oriented updating coordination model, and a hierarchy control framework from a central server to the application-servers.4,Transitions order of runtime-component-state and execution order for short transactions of persistent data conversion included in the long conversion-transaction is automatically produced according to the two kinds of dependence relation between components, generated by composition or service calls. That will ensure the consistency of updated system status data before and after update. On the one hand, it will reduce the burden on developers; on the other hand it will avoid developers'error in arranging the dependence order of conversion transactions.5,With the typed Pi-calculus, type-safety of rebinding service after dynamic updating may be described and verified. This verification is based on the sub-types'match rather than the names'match. It is ensured formally of the correctly rebinding of service and types'safety after the dynamic updating of service configuration and service implementation. Thus it will provide the basis for the updated system consistency before and after update.6,Based on a distribued system BR-OSGi-DS, a prototype implementation supporting dynamic updating is given, which verifies the correctness of relevant strategy from a practical point of view. Experimental method provides a valuable reference for further study of dynamic updating on ubiquitous computing platform supporting ubiquitous service transaction.