Research And Implementation Of High Availability Technology For Distributed Services

With the rapid development of information technology, distributed computing has become one of the mainstream aspects of computer techniques and software has developed gradually from integrated forms to distributed forms. CORBA, one of the mainstream distributed computing specifications, has been applied widely, the amount of service instances is becoming larger and larger and relationships among objects are becoming more and more complex. The services of Massive Data Management Platform (MDMP), which are developed with CORBA product, ofen fail when processing massive data. How to improve the availability of services in MDMP has become a challenge in front of us. Nowadays, there are lots of high availability techniques, but research about availability of mid-ware for massive data processing is very rare.On the base of researching and analyzing existing high availability techniques, this paper designs a series of high availability mechaniques for distributed services which is used for massive data processing and ensure the availability of distributed services from five aspects. This paper adapts three methods to prolong the average working time of services and two techniques to shorten the time for repairing. From the programs of services, this paper improves availability of services by using CORBA Policy and writing running logs to shorten the time for discovering failures of services; from the perspective of services deployment, this paper adds redundant services to the minimum service supporting collection to avoid the unavailability of services when one service breaks down; service-bridge layer adapts redundant service group to improve the robustness of the object references managing service. There are ususlly many instances of distriuted services and it is difficult to maintain them by hand.This paper designs an auto-manage algorithms and improves it, making the process of auto-managing more efficiency, more quickly; This paper improves of existing log alerting techniques, adapting multi-thread mechanics, and adding offline-log detecting and transferring thread to avoid the problem of exhausting disk space.The techniques have been applied to a practical system MDMP, and have a very good effect on improving the availability and performance of distributed services by testing and practical running on line.