Validation of burst scale decay rate and analysis of lognormal delay distribution in NxOn-Off/D/1 queueing models for wireless network performance


Enisoglu R., İNAN B. A., UĞURLU S. Y.

Cluster Computing, cilt.29, sa.13, 2026 (SCI-Expanded, Scopus)

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 29 Sayı: 13
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1007/s10586-026-06591-y
  • Dergi Adı: Cluster Computing
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, Applied Science & Technology Source, Compendex, INSPEC, Academic Search Ultimate (EBSCO), Technology Collection (ProQuest)
  • Anahtar Kelimeler: Burst scale decay rate, Goodness-of-fit testing, Lognormal delay distribution, On-Off/D/1 queueing model, QoS provisioning, Sensitivity analysis, Tail probability estimation, Wireless network dimensioning
  • Akdeniz Üniversitesi Adresli: Evet

Özet

Understanding packet delay behavior in communication networks is essential for Quality of Service (QoS) provisioning and Quality of Experience (QoE) optimization. This paper presents the design, implementation, and validation of an On-Off/D/1 discrete-time queueing simulator that captures the bursty nature of network traffic, together with a statistically rigorous characterization of its queueing-delay distribution. The framework is validated in two stages: the M/D/1 baseline reproduces the exact asymptotic decay rate to within over loads –0.95, and the On-Off/D/1 model tracks the closed-form Burst Scale Decay Rate (BSDR) approximation within 2.2–, where replication-based confidence intervals show that the residual gap is a systematic property of the closed-form approximation rather than simulation noise. The per-packet queueing delay is then analyzed over three ON/OFF configurations and loads –0.98, using five independent replications and approximately packets per operating point. Four distance measures (Kolmogorov–Smirnov, Anderson–Darling, CDF coefficient of determination, Bhattacharyya overlap) consistently show that the delay distribution approaches a lognormal form as the load increases, with the KS distance shrinking to 0.051 and the fixed-sample Anderson–Darling statistic decreasing three- to seven-fold; bootstrap-corrected goodness-of-fit tests nevertheless reject exact lognormality at every operating point, so the lognormal is presented as an approximation with quantified accuracy rather than an exact law. A systematic comparison against Weibull, gamma, log-logistic, and Lomax alternatives shows that the lognormal becomes competitive with the best two-parameter fits as the load increases. Sensitivity analysis shows that OFF-dominated or symmetric configurations () yield better lognormal agreement at moderate loads, with the dependence weakening near saturation. Finally, tail percentile analysis shows that above the fitted lognormal consistently overestimates the empirical P95/P99/P99.9 delays in all tested configurations—by factors of up to 3.6– at P99.9—so that it acts as an empirically conservative closed-form tool for delay budgeting in the heavily loaded regime, whereas at moderate loads the sign of the tail error depends on the traffic configuration. These findings provide uncertainty-quantified guidance for network dimensioning and delay-sensitive traffic engineering.