关键词: 公共对象请求代理体系结构, 容错机制, 对象冗余, 网络分割, 复制对象, 分布式系统

Abstract: Fault-Tolerant Common Object Request Broker Architecture (FT-CORBA) released by Object Management Group (OMG) provides fault-tolerance by replication. However, one of the main drawbacks of the FT-CORBA is that it does not support network partition, which limits the availability of the system severely since it makes the system access the states of adequate replications again and again. By adding appendix to the ORB middleware, classifying the requirement of the replication's uniformity into two levels, and acting according to the different levels, distributed system can better perform during partition failure. Considering the situation that network has been partitioned into 3 and 10 subnets: when there are three subnets, improved replication mechanism has pretty good performance no matter how many notes the subnets contain; when the network is divided into ten subnets, the availability of the distributed system is mostly dependent on the number of the nodes that a single subnet contains. If no subset contains most nodes of the distributed system, the availability also stays low but still higher than before. Since solution is applied to an off-the-set ORB, there is no need to have ORB's source code modified.

Key words: Common Object Request Broker Architecture (CORBA), fault tolerance, entity redundancy, network partition, object replication, distributed system