The Storage Hierarchy is Not a Hierarchy: Optimizing Caching on Modern Storage Devices...
summary
Proposes non‑hierarchical caching (NHC) implemented as Orthus‑CAS (block‑layer kernel module) and Orthus‑KV (user‑level for a key‑value store) that add a tunable data‑admit flag and a load switch (node‑admit ratio) plus a feedback scheduler that optimizes any user‑provided metric f (throughput, average, p99). NHC offloads misses to slower devices when cache is saturated, adapts at runtime, and shows up to 2x throughput gains over classic caching across real device pairs and simulated hierarchies.
AI-written summary · about
chapters (12)
- 0:00Intro
- 0:33Storage Hierarchy
- 1:14Caching Wisdom: Maximizing Hit Rates
- 2:02The Modern Storage Hierarchies
- 4:17Augmenting Caching with Dynamic Load Admission and Request offloading • Intuition: we should avoid excess load to cache device when it is saturated
- 5:04Design: Non-Hierarchical Caching
- 6:12How much load to offload?
- 7:21Example NHC Scheduling States
- 8:57NHC - Key Properties
- 9:31Implementation
- 9:59Ability to utilize capacity device performance
- 11:27Other Experiments in the Paper
related talks
19mP2CACHE: Exploring Tiered Memory for In-Kernel File Systems Caching
43mWhere Is My Cache? Architectural Patterns for Caching ...
25mArchitectural Caching Patterns for Kubernetes
47mCaching beyond simple put and gets
09mCache 101: Understand key strategies before applying them
43mCache Me Outside - Caching Methodologies & Architectures
26mCache Strategies with Best Practices
55mWriting Cache-Friendly C++