Multi-Ring Paxos

Abstract

This paper addresses the scalability of group communication protocols. Scalability has become an issue of prime importance as data centers become commonplace. By scalability we mean the ability to increase the throughput of a group communication protocol, measured in number of requests ordered per time unit, by adding resources (i.e., nodes). We claim that existing group communication protocols do not scale in this respect and introduce Multi-Ring Paxos, a protocol that orchestrates multiple instances of Ring Paxos in order to scale to a large number of nodes. In addition to presenting Multi-Ring Paxos, we describe a prototype of the system we have implemented and a detailed evaluation of its performance.

HT-Ring Paxos: Theory of High Throughput State-Machine Replication for Clustered Data Centers

“…Multi-Ring Paxos [27] uses the concept of State partitioning [8] for achieve scalability. Multi-Ring Paxos have various logical groups.…”

Scalable State-Machine Replication

“…To do this, a replica receives the request, executes it locally and then communicates with all other replicas in the system to check if there are conflicting requests being executed at the same time. If not, then the request execution is confirmed and the state of the replicas is changed accordingly; otherwise, it is cancelled and the replica state remains unchanged (PACHECO;SCHIPER, 2003;PEDONE, 2012;PEDONE, 2013;JUNQUEIRA, 2012). In either case, the client is notified about the result (Figure 3.4).…”

D-Paxos: Building Hierarchical Replicated State Machine for Cloud Environments

“…Multi-Ring Paxos [20] uses multiple instances of Ring Paxos [21] to order all requests in which those instances have an interest and thus scale throughput without sacrificing response time. Multi-Ring Paxos behaves similarly to D-Paxos, because learners subscribing to multiple groups of Ring Paxos instances should also combine multiple preordered sequences of requests from them.…”