This paper relies on 2 prerequisite mathematical and architectural foundations.
Paxos and Fast Paxos are optimal consensus algorithms that are simple and elegant, while suitable for efficient implementation. In this paper, we compare the performance of both algorithms in failure-free and failure-prone runs using Treplica, a general replication toolkit that implements these algorithms in a modular and efficient manner. We have found that Paxos outperforms Fast Paxos for small number of replicas and that collisions are not the cause of this performance difference.
Addresses critical efficiency and scalability bottlenecks in modern computing systems, providing concrete empirical evidence and reproducible designs.
In this paper, we compare the performance of both algorithms in failure-free and failure-prone runs using Treplica, a general replication toolkit that implements these algorithms in a modular and efficient manner.
Paxos and Fast Paxos are optimal consensus algorithms that are simple and elegant, while suitable for efficient implementation.
We have found that Paxos outperforms Fast Paxos for small number of replicas and that collisions are not the cause of this performance difference.
Trade-offs between peak accuracy and memory overhead remain under investigation across diverse hardware environments.
Consensus protocols are crucial for reliable distributed systems as they let them cope with network and server failures. For decades, most consensus protocols have been designed as variations of the seminal Paxos, yet in 2014 Raft was presented as a new, "understandable" protocol, meant to be easier to implement than the notoriously subtle Paxos family. Raft has since been used in various industrial projects, e.g. Hashicorp's Consul or etcd (used by Google's Kubernetes). The correctness of Raft is established via a manual proof, based on a TLA+ specification of the protocol. This paper reports our experience in modeling Raft in the LNT process algebra. We found a couple of issues with the original TLA+ specification of Raft, which has been corrected since. More generally, this exercise offers a great opportunity to discuss how to best use the features of the LNT formal language and the associated CADP verification toolbox to model distributed protocols, including network and server failures.
Distributed consensus is a fundamental primitive for constructing fault-tolerant, strongly-consistent distributed systems. Though many distributed consensus algorithms have been proposed, just two dominate production systems: Paxos, the traditional, famously subtle, algorithm; and Raft, a more recent algorithm positioned as a more understandable alternative to Paxos. In this paper, we consider the question of which algorithm, Paxos or Raft, is the better solution to distributed consensus? We analyse both to determine exactly how they differ by describing a simplified Paxos algorithm using Raft's terminology and pragmatic abstractions. We find that both Paxos and Raft take a very similar approach to distributed consensus, differing only in their approach to leader election. Most notably, Raft only allows servers with up-to-date logs to become leaders, whereas Paxos allows any server to be leader provided it then updates its log to ensure it is up-to-date. Raft's approach is surprisingly efficient given its simplicity as, unlike Paxos, it does not require log entries to be exchanged during leader election. We surmise that much of the understandability of Raft comes from the paper's clear presentation rather than being fundamental to the underlying algorithm being presented.
