Work

Using Lock-Free Design for Throughput-Optimized Cache Eviction

Year: 2025

Type: article

Abstract: This paper presents a practical approach to cache eviction algorithm design, called Mobius, that optimizes the concurrent throughput of caches and reduces cache operation latency by utilizing lock-free data structures, while maintaining high cache hit ratios. Mobius includes two key designs. First, Mobius employs two lock-free FIFO queues to manage cache items, ensuring that all cache operations are executed efficiently in concurrency. Second, Mobius integrates a consecutive detection mechanism that merges multiple modifications during eviction into a single operation, thereby reducing data races. The implementation of Mobius in CacheLib and RocksDB highlights its high concurrency in both synthetic and real-world workloads.

Source: ACM SIGMETRICS Performance Evaluation Review

DOI: 10.1145/3744970.3727330

Authors: Chao Dong, Fang Wang, Hong Jiang, Dan Feng

Institutions: Wuhan National Laboratory for Optoelectronics, Huazhong University of Science and Technology, The University of Texas at Arlington

Language: English

Topic: Parallel Computing and Optimization Techniques

Subfield: Hardware and Architecture

Field: Computer Science

Domain: Physical Sciences

Open Access status: bronze

Funders: National Natural Science Foundation of China, Universitas Brawijaya, National Key Research and Development Program of China