{
  "aliases": [],
  "status": "research",
  "supersedes": [],
  "supersededBy": [],
  "firstSeen": null,
  "lastSeen": null,
  "updated": "2026-11-19",
  "slug": "ioflow",
  "term": "IOFlow",
  "definition": "IOFlow is a research architecture that exposes storage-stage queues to a logically centralized controller for end-to-end I/O policies.",
  "note": "The controller maps high-level flows onto stage queues with service and routing controls. Storage tokens represent operation cost, not simply request-header bytes. [ioflow-paper]\n\nIts Windows prototype inserts drivers around SMB stages without requiring application or VM changes. Requests already inside an uncontrolled device remain outside its priority decisions; limiting outstanding work retains more scheduling choice upstream. [ioflow-paper]\n\nFor evaluation, name the enforcement stage, operation-cost model and outstanding-request boundary. A configured rate alone does not establish application tail latency.",
  "related": [
    "sroute",
    "io-scheduler",
    "mclock",
    "queue-depth"
  ],
  "citations": [
    {
      "id": "ioflow-paper",
      "title": "Thereska et al. IOFlow: A Software-Defined Storage Architecture. SOSP 2013, sections 3.1, 4.1 and 5; DOI 10.1145/2517349.2522723.",
      "url": "https://www.microsoft.com/en-us/research/wp-content/uploads/2013/11/ioflow-sosp13.pdf",
      "accessed": "2026-11-19"
    }
  ],
  "advantages": [
    {
      "text": "A controller can coordinate policies across several storage stages.",
      "citationIds": [
        "ioflow-paper"
      ]
    }
  ],
  "limitations": [
    {
      "text": "Priority enforcement depends on where contention occurs and which stages expose control.",
      "citationIds": [
        "ioflow-paper"
      ]
    }
  ],
  "provenance": [
    "Primary-paper architecture review; no prototype execution or contemporary performance measurement."
  ]
}
