Pineapple - Unifying Multi-Paxos and Atomic Shared Registers

Explore a conference presentation that introduces Pineapple, a novel distributed systems approach that unifies Multi-Paxos consensus algorithms with atomic shared registers to improve linearizable storage system performance. Learn how this innovative design uses logical timestamps to enable any node in the system to serve both read and write operations, effectively reducing bottlenecks typically associated with leader-based consensus systems. Discover the technical advantages of Pineapple's "non-blocking operation execution" compared to existing systems like Gryff, which combines EPaxos with atomic shared registers. Examine comprehensive evaluation results demonstrating significant performance improvements in both throughput and tail latency across wide-area and local-area networks when compared to state-of-the-art systems including Gryff, Multi-Paxos, and EPaxos. Understand the practical implementation through the integration with etcd, where Pineapple achieves over 50% reduction in median latency compared to optimized Raft-based systems in balanced workloads. Gain insights into how this approach addresses the complexity challenges of developing correct large-scale customer-facing applications in the presence of concurrent operations and system failures.