基于CORBA消息服务的容错机制研究

CORBA逐渐成为面向对象分布式应用中间件的主要标准，但是CORBA当前没有为容错提供相应的机制，该文在比较各种容错方法的基础上，讨论了一种基于异步消息服务回调和查询模型的容错方法，该方法使服务对象的副本可以并行地处理客户请求，提高了容错的性能，该方法具有对服务对象透明，可以满足用户不同容错要求等特点，文中还给出了这种方法在为遗留应用增加容错能力时的应用实例。

Fault-Tolerant Mechanism Basing on Corba Messaging

This paper provides a new means of solution based on asynchronous messaging service callback and polling model in CORBA environment. In messaging callback model, clients can send requests in non-blocking way, without waiting for server-side replies. When the server finishes processing, notifying events be sent to client registered handle object by callbacks. To deal with fault-tolerant with callbacks, clients send each request to all server replicas simultaneously. Reply handle object for each server replica is registered correspondingly. The client only responds to the first server reply and discards rest latecomers. In CORBA message polling model, clients can send unblocking requests and quest for result at will. Take the fault-tolerant into account, one client sends each request to all server replicas, inquires for reply in round-robin way. Once a reply be obtained, no more polling work needed. The callback model works in event-driven style, which means no busy stall. More effective system resource utilization could be achieved. Separated from application logic, fault-tolerant implementations are integrated into asynchronous reply objects. Such structure brings more manageability and maintainability. It also results in complex control logic, suitable for applications with complicated fault-tolerant demands and large number of server objects. The message-polling model acts in traditional way, according with legacy programming style. Round-robin queries may bring system stall, resulting in overall performance descends. Under such circumstance, fault-tolerant logic is tightly coupled with application object, which brings difficulty to system management. Such coupled integration also engenders simplicity in control mechanism design and implementation, and is suitable for ordinary fault-tolerant requirements with a small quantity of server objects. The approach the paper presents enable replicas process request in parallel, thus better fault-tolerant performance could be expected. Other benefits include server-side object transparency, which means more adaptability to various customer demands. Conclusively, the paper shows an example of applying the approach discussed above for legacy applications to improve fault-tolerant capacity.