[Paper] Highly Reliable Redundant Data Allocation Method for High Availability Distributed Clusters

抄録

In mission-critical systems such as session control servers provided by telecommunications carriers, we are researching distributed cluster architectures that enable flexible control of system performance. Data must be maintained even in the event of a disaster. Considering the physical distance between the servers that manage the original and duplicate data respectively, we proposed a data redundancy control method that does not cause data loss even in the event of a large-scale disaster. The proposed method was evaluated by computer simulation. We also proposed an implementation of the data loss avoidance method in Apache Cassandra that defines “Racks” as a “site”. We implemented the method and confirmed that site collision avoidance was achieved without data loss even in the case of a large-scale disaster. We found that the number of servers in each site should be about the same for optimal data storage in terms of decentralization and reliability.