GRADE: Graceful Degradation in Byzantine Quorum Systems

Abstract

Distributed storage systems are expected to provide correct services in the presence of Byzantine failures, which do not have any assumptions about the behavior of faulty servers and clients. In designing such systems, we often encounter the paradox of fault tolerance vs. performance (or efficiency), because better fault tolerance usually requires a tradeoff of system performance. In this paper, we present GRADE, a Byzantine-fault-tolerant (BFT) distributed storage system that enables graceful degradation. Two Byzantine quorum systems (BQSs) are supported on each GRADE server: a masking BQS storing generic data and a dissemination BQS storing self-verifying ones. Based on the system status and the environment, servers dynamically and seamlessly switch between two BQSs, without converting the stored data. Therefore, GRADE provides high performance in a normal running-state, and degrades performance to maintain high fault tolerance in emergency situations. The computation and communication costs of the running-state switch are very low, and the switch is completely transparent to clients. Our performance analysis and experimental results demonstrate that GRADE provides a balance between performance and fault tolerance.