Lune

FAST2024Top-tier venue

Baleen: ML Admission & Prefetching for Flash Caches

Daniel Lin-Kit Wong, Hao Wu, Carson Molder, Sathya Gunasekar, Jimmy Lu, Snehal Khandkar, Abhinav Sharma, Daniel S. Berger, Nathan Beckmann, Gregory R. Ganger

Abstract

Flash caches are used to reduce peak backend load for throughput-constrained data center services, reducing the total number of backend servers required. Bulk storage systems are a large-scale example, backed by high-capacity but lowthroughput hard disks, and using flash caches to provide a more cost-effective storage layer underlying everything from blobstores to data warehouses.

However, flash caches must address the limited write endurance of flash by limiting the long-term average flash write rate to avoid premature wearout. To do so, most flash caches must use admission policies to filter cache insertions and maximize the workload-reduction value of each flash write.

The Baleen flash cache uses coordinated ML admission and prefetching to reduce peak backend load. After learning painful lessons with our early ML policy attempts, we exploit a new cache residency model (which we call episodes) to guide model training. We focus on optimizing for an end-toend system metric (Disk-head Time) that measures backend load more accurately than IO miss rate or byte miss rate. Evaluation using Meta traces from seven storage clusters shows that Baleen reduces Peak Disk-head Time (and hence the number of backend hard disks required) by 12% over stateof-the-art policies for a fixed flash write rate constraint. Baleen-TCO, which chooses an optimal flash write rate, reduces our estimated total cost of ownership (TCO) by 17%. Code and traces are available via https://www.pdl.cmu.edu/CILES/.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

11 top-tier papers have cited it since. Know what they found before your reviewers do.

Questions to start from

Lune can draw on

the full paper · the top-tier papers that cite it · related work

Get the answer

Free to start. No credit card required.

Lineage

  • Heimdall: Optimizing Storage I/O Admission with Extensive Machine Learning Pipeline cites Facebook's Tectonic Filesystem: Efficiency from Exascale
  • Reducing Cross-Cloud/Region Costs with the Auto-Configuring MACARON Cache cites Learning Relaxed Belady for Content Distribution Network Caching
  • DumpKV: Learning based lifetime aware garbage collection for key value separation in LSM-tree cites Learning Relaxed Belady for Content Distribution Network Caching
  • The CacheLib Caching Engine: Design and Experiences at Scale cites Learning Relaxed Belady for Content Distribution Network Caching
  • Facebook's Tectonic Filesystem: Efficiency from Exascale cites The CacheLib Caching Engine: Design and Experiences at Scale
  • Towards Efficient Flash Caches with Emerging NVMe Flexible Data Placement SSDs cites The CacheLib Caching Engine: Design and Experiences at Scale
Builds onCited byLeft to right by year; the more cited, the higher.

Cited by top-tier papers11

  1. Towards Efficient Flash Caches with Emerging NVMe Flexible Data Placement SSDsMichael Allison, Arun George, Javier González, Dan Helmick et al.EuroSys 2025 · 13 citations
  2. Heimdall: Optimizing Storage I/O Admission with Extensive Machine Learning PipelineDaniar Heri Kurniawan, Rani Ayu Putri, Peiran Qin, Kahfi S. Zulkifli et al.EuroSys 2025 · 3 citations
  3. Reducing Cross-Cloud/Region Costs with the Auto-Configuring MACARON CacheHojin Park, Ziyue Qiu, Gregory R. Ganger, George AmvrosiadisSOSP 2024 · 3 citations
  4. DumpKV: Learning based lifetime aware garbage collection for key value separation in LSM-treeZhutao Zhuang, Zhiguang Chen, Xinqi ZengVLDB 2025 · 3 citations
  5. Here, There and Everywhere: The Past, the Present and the Future of Local Storage in CloudLeping Yang, Yanbo Zhou, Gong Zeng, Li Zhang et al.FAST 2026 · 2 citations

Builds on15

  1. Learning Relaxed Belady for Content Distribution Network CachingZhenyu Song, Daniel S. Berger, Kai Li, Wyatt LloydNSDI 2020 · 193 citations
  2. The CacheLib Caching Engine: Design and Experiences at ScaleBenjamin Berg, Daniel S. Berger, Sara McAllister, Isaac Grosof et al.OSDI 2020 · 145 citations
  3. Facebook's Tectonic Filesystem: Efficiency from ExascaleSatadru Pan, Theano Stavrinos, Yunqiao Zhang, Atul Sikaria et al.FAST 2021 · 110 citations
  4. An Imitation Learning Approach for Cache ReplacementEvan Zheran Liu, Milad Hashemi, Kevin Swersky, Parthasarathy Ranganathan et al.ICML 2020 · 108 citations
  5. OSCA: An Online-Model Based Cache Allocation Scheme in Cloud Block Storage SystemsYu Zhang, Ping Huang, Ke Zhou, Hua Wang et al.USENIX ATC 2020 · 74 citations

Related papers

  1. FairyWREN: A Sustainable Cache for Emerging Write-Read-Erase Flash InterfacesSara McAllister, Yucong Wang, Benjamin Berg, Daniel S. Berger et al.OSDI 2024 · 14 citations
  2. Nemo: A Low-Write-Amplification Cache for Tiny Objects on Log-Structured Flash DevicesXufeng Yang, Tingting Tan, Jingxin Hu, Congming Gao et al.ASPLOS 2026
  3. Making Cache Monotonic and ConsistentShuai An, Yang CaoVLDB 2023 · 1 citation
  4. Your Read is Our Priority in Flash StorageMijin An, Soojun Im, Dawoon Jung, Sang Won LeeVLDB 2022 · 9 citations
  5. CDCache: Space-Efficient Flash Caching via Compression-before-DeduplicationHengying Xiao, Jingwei Li, Yanjing Ren, Ruijin Wang et al.INFOCOM 2024 · 1 citation
Dusk over the sea between two cliffs drawn in fine vertical lines