Steady-state GI/G/\(n\) queue in the Halfin-Whitt regime (Q389068): Difference between revisions

From MaRDI portal
Importer (talk | contribs)
Changed an Item
Added link to MaRDI item.
links / mardi / namelinks / mardi / name
 

Revision as of 03:18, 30 January 2024

scientific article
Language Label Description Also known as
English
Steady-state GI/G/\(n\) queue in the Halfin-Whitt regime
scientific article

    Statements

    Steady-state GI/G/\(n\) queue in the Halfin-Whitt regime (English)
    0 references
    0 references
    0 references
    17 January 2014
    0 references
    The paper studies the first-come-first-served (FCFS) GI/GI/\(n\) queueing system in the Halfin-Whitt regime, see [\textit{S. Halfin} and \textit{W. Whitt}, Oper. Res. 29, 567--588 (1981; Zbl 0455.60079)]. The paper proves the tightness of the sequence of steady state queue-length distributions, normalized by the \(\sqrt{n}\), as \(n\to\infty\). The paper derives an upper bound of the large deviation exponent of the limiting steady state queue-length matching. Earlier, this upper bound was conjectured in [\textit{D. Gamarnik} and \textit{P. Momčilović}, Adv. Appl. Probab. 40, No. 2, 548--577 (2008; Zbl 1148.60070)]. Under the assumption that the arrival process is Poisson, a matching lower bound is proved as well. The paper derives new and simple bounds on the basis of new techniques for the FCFS GI/GI/\(n\) queueing system. The bounds are of a structural nature and hold for all \(n\) and all times \(t\geq 0\). The closed form representations for these bounds are intuitively explained as the suprema of certain natural processes weakly converging to Gaussian processes. The aforementioned weak convergence was earlier studied in [\textit{J. Reed}, Ann. Appl. Probab. 19, No. 6, 2211--2269 (2009; Zbl 1181.60137)]. The methodology of the present paper establishes the first non-trivial bound for the aforementioned weak limit process in [Reed, loc. cit.].
    0 references
    many-server queues
    0 references
    large deviations
    0 references
    weak convergence
    0 references
    Gaussian process
    0 references
    stochastic comparison
    0 references

    Identifiers