Byz-GentleRain: An Efficient Byzantine-tolerant Causal Consistency Protocol

Abstract

Causal consistency is a widely used weak consistency model that allows high availability despite network partitions. There are plenty of research prototypes and industrial deployments of causally consistent distributed systems. However, as far as we know, none of them consider Byzantine faults, except Byz-RCM proposed by Tseng et al. Byz-RCM achieves causal consistency in the client-server model with 3f + 1 servers where up to f servers may suffer Byzantine faults, but assumes that clients are non-Byzantine. In this work, we present Byz-Gentlerain, the first causal consistency protocol which tolerates up to f Byzantine servers among 3f + 1 servers in each partition and any number of Byzantine clients. Byz-GentleRain is inspired by the stabilization mechanism of GentleRain for causal consistency. To prevent causal violations due to Byzantine faults, Byz-GentleRain relies on PBFT to reach agreement on a sequence of global stable times and updates among servers, and only updates with timestamps less than or equal to such common global stable times are visible to clients. We prove that Byz-GentleRain achieves Byz-CC, the causal consistency variant in the presence of Byzantine faults. We evaluate Byz-GentleRain on Aliyun. The preliminary results show that Byz-GentleRain is efficient on typical workloads.

Byzantine-Tolerant Causal Ordering for Unicasts, Multicasts, and Broadcasts

“…Recently there has been some work on the related problem of implementing Byzantine-tolerant causal consistency in distributed shared memory and replicated databases [18], [19], [20]; these approaches relied on broadcast communication. In [18], Byzantine Reliable Broadcast (BRB) [21] is used to remove misinformation induced by the combination of asynchrony and Byzantine behaviour.…”

“…In [18], Byzantine Reliable Broadcast (BRB) [21] is used to remove misinformation induced by the combination of asynchrony and Byzantine behaviour. In [19], PBFT (total order broadcast) [22] is used to achieve consensus among non-Byzantine servers regarding the order of client requests. In [20], Byzantine causal broadcast has been used to implement Byzantine eventual consistency.…”

Byzantine Fault Tolerant Causal Ordering

“…We show that a matrix clock based data structure is required for ordering point-to-point messages and therefore the broadcast technique will not be applicable to a point-to-point setting. There has been recent interest in applying the Byzantine fault model in implementing causal consistency in distributed shared memory and replicated databases [7], [5], [6]. In [6], Byzantine causal broadcast has been used to implement Byzantine eventual consistency.…”

“…induced by the combination of asynchrony and Byzantine behaviour. In [5], PBFT (total order broadcast) [16] is used to achieve consensus among non-Byzantine servers regarding the order of client requests. To the best of our knowledge, no paper has attempted to solve causal ordering of point-to-point messages and multicasts in an asynchronous system with Byzantine failures.…”

“…Most of the work on causal ordering of messages did not consider the Byzantine failure model. The only work on causal ordering under the Byzantine failure model was the recent result by Auvolat et al [4] which considered only Byzantinetolerant causal broadcasts, and the work in [5], [6], [7] which relied on broadcasts. To the best of our knowledge, there has been no work on Byzantine-tolerant causal ordering of pointto-point messages.…”