Reliable Replication at Low Cost

The emerging global scientific collaborations demand a scalable, efficient, reliable, and still convenient data access and management scheme. To fulfill these requirements, this paper describes a replicated file system that supports mutable (i.e., read/write) replication with strong consistency guarantees, small performance penalty, high failure resilience, and good scaling properties. The paper further evaluates the system using a real scientific application. The evaluation results show that the presented replication system can significantly improve the application’s performance by reducing the first-time access latency to read the input data and by distributing the verification of data access to a nearby server. Furthermore, the penalty of file replication is negligible as long as applications use synchronous writes at a moderate rate.