Reading: Related Work Vocabulary — English for Researchers

2. Related Work

Our work builds on the foundational line of research on adaptive cache replacement policies established by Smith et al. [12], extending their single-tier model to multi-tier caching hierarchies. A separate body of prior art has explored machine-learning-based eviction policies [7, 15, 22]; however, these approaches typically require offline training on representative workloads and do not adapt online, which limits their applicability to workloads that shift over time. Concurrently with our work, Chen et al. [31] independently proposed a similar dynamic quorum mechanism, though their evaluation focuses exclusively on read-heavy workloads, whereas we additionally evaluate write-heavy and mixed workloads. To our knowledge, no prior work has combined online adaptation with formal consistency guarantees, which is the gap our approach addresses. We refer readers to Ousterhout et al. [3] for a broader survey of the distributed caching landscape.