Efficient replication via timestamp stability

Abstract

Modern web applications replicate their data across the globe and require strong consistency guarantees for their most critical data. These guarantees are usually provided via statemachine replication (SMR). Recent advances in SMR have focused on leaderless protocols, which improve the availability and performance of traditional Paxos-based solutions. We propose Tempo -a leaderless SMR protocol that, in comparison to prior solutions, achieves superior throughput and offers predictable performance even in contended workloads. To achieve these benefits, Tempo timestamps each application command and executes it only after the timestamp becomes stable, i.e., all commands with a lower timestamp are known. Both the timestamping and stability detection mechanisms are fully decentralized, thus obviating the need for a leader replica. Our protocol furthermore generalizes to partial replication settings, enabling scalability in highly parallel workloads. We evaluate the protocol in both real and simulated geo-distributed environments and demonstrate that it outperforms state-of-the-art alternatives.CCS Concepts: • Theory of computation → Distributed algorithms.

Network Bandwidth Variation-Adapted State Transfer for Geo-Replicated State Machines and its Application to Dynamic Replica Replacement

“…Geographic SMR places replicas at a large distance from each other and tolerates large-scale disasters, such as earthquakes. Since geographic SMR exhibits different characteristics from conventional SMR, there are various SMR protocols specialized for geographic SMR [2,3,10,11,14,16,15,18,17,24,25,29,30,36,38,39,40]. The most significant difference between SMR in a data center and geographic SMR is the inequality of latency and throughput between replicas and clients.…”

“…However, since the SMR protocols designed for a data center assume that messages arrive at almost the same time on all replicas, these SMR protocols cannot demonstrate their original performance in geographic SMR. Therefore, many SMR protocols designed for geographic SMR [25, 40, 36, Transfer replica 3,24,14,10,30,11,39,15,18,17] assume different latencies and use various techniques to overcome this difference.…”

“…2.1, many SMR protocols require a leader to order the requests, in which replicas must forward the requests to the leader. Since this transfer adversely affects latency, geographic SMR often uses leaderless protocols [15,18,17]. In a leaderless protocol, any replica can determine the processing order of requests by coordinating with other replicas.…”

“…Furthermore, performing SMR with multiple geographically separated replicas allows the service to resist large-scale disasters, such as earthquakes. Such a method is called geographic SMR [2,3,10,11,14,16,15,18,17,24,25,29,30,36,38,39,40].…”

Regular Sequential Serializability and Regular Sequential Consistency

“…Gryff is a recent geo-replicated storage system that combines shared registers and consensus [15]. Many other protocols have been developed to provide replicated and linearizable storage [3,27,28,30,38,44,62,64,67,68,92]. Weakening the consistency of these other services to RSC is likely to enable new variants of their designs that improve their performance.…”