{
  "slug": "kosmo",
  "term": "Kosmo",
  "aliases": [
    "Kosmo MRC estimator"
  ],
  "definition": "Kosmo estimates miss-ratio curves for supported cache-eviction policies using shared object records and policy-specific eviction maps.",
  "note": "Mechanism: eviction maps reconstruct cache states when needed; a histogram of minimum containing-cache sizes becomes the estimated curve. The design also uses SHARDS sampling. [kosmo-paper]\n\nBoundary: the output is nonincreasing even for policies that violate inclusion. It cannot exactly represent an upward segment of the actual miss-ratio curve. [kosmo-paper]\n\nPolicy identity: Kosmo distinguishes ideal LFU, which retains frequency counters for evicted objects, from the modeled reset-on-eviction variant. Reaccess starts the latter counter at one; its resident sets can violate inclusion. A stack-method result for ideal LFU does not automatically apply to this variant. [kosmo-paper]\n\nEvaluation: LFU and FIFO used 52 traces, while 2Q and LRFU used 33. The described LRU and MRU variants were not included in the performance evaluation. [kosmo-paper]\n\nAlternative: Miniature Simulations evaluates sampled caches at selected capacities; it does not impose the same single-histogram shape. It still has sampling error and configuration costs. [minisim-paper]\n\nExact illustration: the existing FIFO example has 9/12 misses at three slots and 10/12 at four. Any nonincreasing estimate must have maximum absolute error at least 1/24 over those two points. This is an algebraic bound, not a measured Kosmo error or a bound on mean error. [hesela-bound]",
  "status": "research",
  "advantages": [],
  "limitations": [],
  "related": [
    "miss-ratio-curve",
    "cache-inclusion-property",
    "belady-anomaly",
    "fifo-replacement"
  ],
  "supersedes": [],
  "supersededBy": [],
  "firstSeen": null,
  "lastSeen": null,
  "updated": "2026-11-01",
  "citations": [
    {
      "id": "kosmo-paper",
      "title": "Shakiba, Sultan and Stumm. Kosmo: Efficient Online Miss Ratio Curve Generation for Eviction Policy Evaluation. FAST 2024.",
      "url": "https://www.usenix.org/system/files/fast24-shakiba.pdf",
      "accessed": "2026-11-01"
    },
    {
      "id": "minisim-paper",
      "title": "Waldspurger, Saemundson, Ahmad and Park. Cache Modeling and Optimization using Miniature Simulations. USENIX ATC 2017, pp. 487-498; sections 1.1 and 3.1-3.2.",
      "url": "https://www.usenix.org/system/files/conference/atc17/atc17-waldspurger.pdf",
      "accessed": "2026-11-01"
    },
    {
      "id": "hesela-bound",
      "title": "Hesela: exact monotone approximation bound from the existing FIFO capacity counts; code and assumptions linked in the companion article.",
      "url": "https://hesela.dev/research/monotone-mrc.json",
      "accessed": "2026-11-01"
    }
  ],
  "provenance": [
    "Source review and exact arithmetic only. No Kosmo implementation was executed."
  ]
}