{
  "slug": "kingman-approximation",
  "term": "Kingman approximation",
  "aliases": [
    "Kingman's formula",
    "VUT approximation"
  ],
  "definition": "The Kingman approximation estimates mean waiting in a single-server FCFS queue from utilization, mean service time and arrival and service variability.",
  "note": "Write E[Q] approximately equal to ((c_a^2 + c_s^2)/2) * rho/(1-rho) * E[S], where rho = lambda E[S], c_a^2 = Var(A)/E[A]^2 and c_s^2 = Var(S)/E[S]^2. A is an interarrival interval, not a request count. The inputs are squared coefficients of variation (SCVs), not unsquared CVs. [practitioners]\n\nThe usual GI/G/1 interpretation has iid interarrival intervals, iid service times independent of arrivals, one work-conserving nonpreemptive FCFS server, unlimited waiting space, finite second moments and 0 < rho < 1. The heavy-traffic basis concerns a family of queues approaching saturation; it does not certify an arbitrary finite-load estimate or a storage p99. Stronger moment-convergence results need their own regularity conditions. [transform]\n\nSCV one does not establish Poisson arrivals. Intervals 0.5 and 3 with probabilities 0.8 and 0.2 have mean one and variance one but are not exponential. Even with equal arrival means and SCVs, exact GI/M/1 mean waits can differ: at 50% utilization and exponential mean-1 ms service, three declared arrival laws give 1, 0.958912 and 0.699539 ms. [hesela-gim1]\n\nAt mean service 1 ms and service SCV one, changing arrival SCV from 0.25 to 4 changes the estimate from 2.5 to 10 ms at 80% utilization. These are evaluations of an approximation, not observed waits or measured approximation errors. Marginal SCVs do not encode serial correlation or feedback from a closed-loop load generator. [hesela-kingman]",
  "status": "research",
  "advantages": [],
  "limitations": [],
  "related": [
    "queueing-delay",
    "service-time",
    "pollaczek-khinchine-formula"
  ],
  "supersedes": [],
  "supersededBy": [],
  "firstSeen": null,
  "lastSeen": null,
  "updated": "2026-11-20",
  "citations": [
    {
      "id": "practitioners",
      "title": "Harchol-Balter and Scully: The most common queueing theory questions asked by computer systems practitioners, section 1; VUT expression and SCV definitions.",
      "url": "https://www.cs.cmu.edu/~harchol/Papers/TeaPACS21.pdf",
      "accessed": "2026-11-19"
    },
    {
      "id": "transform",
      "title": "Boon, Janssen and van Leeuwaarden: Heavy-traffic single-server queues and the transform method, arXiv:2206.09844v1 (2022), sections 1-3.",
      "url": "https://arxiv.org/abs/2206.09844v1",
      "accessed": "2026-11-19"
    },
    {
      "id": "kingman1961",
      "title": "J. F. C. Kingman: The single server queue in heavy traffic, 57(4), 902-904 (1961). Bibliographic provenance; DOI 10.1017/S0305004100036094.",
      "url": "https://doi.org/10.1017/S0305004100036094",
      "accessed": "2026-11-19"
    },
    {
      "id": "hesela-kingman",
      "title": "Hesela: VUT sensitivity calculations and moment-matched non-Poisson counterexample.",
      "url": "https://hesela.dev/research/kingman/README.md",
      "accessed": "2026-11-19"
    },
    {
      "id": "hesela-gim1",
      "title": "Hesela: Equal arrival moments do not identify mean waiting; GI/M/1 model, primary-source equations and rational numerical certificates.",
      "url": "https://hesela.dev/research/gim1-moments/README.md",
      "accessed": "2026-11-19"
    }
  ],
  "provenance": [
    "Source synthesis plus exact arithmetic applied to an approximate formula; no hardware measurements or simulation.",
    "The original 1961 publisher record establishes attribution; the open 2022 manuscript was used to inspect the theory and its stronger assumptions."
  ]
}
