Distributed Locking: Performance Analysis and Optimization Strategies
Andre Rodriguez, William Osborn
TL;DR
This paper investigates the performance of distributed locking in large-scale and geo-distributed systems, comparing centralized lock managers with distributed protocols under varied workloads. It implements CLM, Paxos/Raft-based PDL, LDL, and HL, and evaluates them on a testbed of $64$ nodes across three regions, using realistic traffic patterns. Findings show distributed protocols, particularly LDL, deliver higher throughput under high contention, while CLM is favorable at low contention but suffers tail latency and poor scalability; geo-distribution further emphasizes the benefits of multi-level designs like HL. To close the gap, the authors introduce locality-aware lock placement, adaptive lease duration, and hybrid locking strategies, reporting up to $68\%$ additional performance gains over baseline approaches. The results provide practical performance guidelines for selecting and configuring locking strategies in modern distributed systems.
Abstract
Distributed locking mechanisms are fundamental to ensuring data consistency and integrity in distributed systems. This paper presents a comprehensive analysis of distributed locking algorithms, focusing on their performance characteristics under various workload conditions. We compare traditional centralized locking approaches with modern distributed protocols, evaluating them based on throughput, latency, and scalability metrics. Our experimental results demonstrate that optimized distributed locking protocols can achieve up to 68\% better performance compared to centralized approaches in high-contention scenarios, while maintaining strong consistency guarantees. Furthermore, we propose novel optimizations for distributed locking that significantly reduce coordination overhead in geo-distributed deployments. The findings contribute to the growing body of knowledge on designing efficient concurrency control mechanisms for modern distributed systems.