English

Excessive Backlog Probabilities of Two Parallel Queues

Probability 2018-06-05 v1

Abstract

Let XX be the constrained random walk on Z+2{\mathbb Z}_+^2 with increments (1,0)(1,0), (1,0)(-1,0), (0,1)(0,1) and (0,1)(0,-1); XX represents, at arrivals and service completions, the lengths of two queues working in parallel whose service and interarrival times are exponentially distributed with arrival rates λi\lambda_i and service rates μi\mu_i, i=1,2i=1,2; we assume λi<μi\lambda_i < \mu_i, i=1,2i=1,2, i.e., XX is assumed stable. Without loss of generality we assume ρ1=λ1/μ1ρ2=λ2/μ2\rho_1 =\lambda_1/\mu_1 \ge \rho_2 = \lambda_2/\mu_2. Let τn\tau_n be the first time XX hits the line An={xZ2:x(1)+x(2)=n}\partial A_n = \{x \in {\mathbb Z}^2:x(1)+x(2) = n \}. Let YY be the same random walk as XX but only constrained on {yZ2:y(2)=0}\{y \in {\mathbb Z}^2: y(2)=0\} and its jump probabilities for the first component reversed. Let B={yZ2:y(1)=y(2)}\partial B =\{y \in {\mathbb Z}^2: y(1) = y(2) \} and let τ\tau be the first time YY hits B\partial B. The probability pn=Px(τn<τ0)p_n = P_x(\tau_n < \tau_0) is a key performance measure of the queueing system represented by XX (probability of overflow of a shared buffer during system's first busy cycle). Stability of XX implies pnp_n decays exponentially in nn when the process starts off An.\partial A_n. We show that, for xn=nxx_n= \lfloor nx \rfloor, xR+2x \in {\mathbb R}_+^2, x(1)+x(2)1x(1)+x(2) \le 1, x(1)>0x(1) > 0, P(nxn(1),xn(2))(τ<)P_{(n-x_n(1),x_n(2))}( \tau < \infty) approximates Pxn(τn<τ0)P_{x_n}(\tau_n < \tau_0) with exponentially vanishing relative error. Let r=(λ1+λ2)/(μ1+μ2)r = (\lambda_1 + \lambda_2)/(\mu_1 + \mu_2); for r2<ρ2r^2 < \rho_2 and ρ1ρ2\rho_1 \neq \rho_2, we construct a class of harmonic functions from single and conjugate points on a characteristic surface of YY with which Py(τ<)P_y(\tau < \infty) can be approximated with bounded relative error. For r2=ρ1ρ2r^2 = \rho_1 \rho_2, we obtain Py(τ<)=ry(1)y(2)+r(1r)rρ2(ρ1y(1)ry(1)y(2)ρ1y(2)).P_y(\tau < \infty) = r^{y(1)-y(2)} +\frac{r(1-r)}{r-\rho_2}\left( \rho_1^{y(1)} - r^{y(1)-y(2)} \rho_1^{y(2)}\right).

Keywords

Cite

@article{arxiv.1806.00686,
  title  = {Excessive Backlog Probabilities of Two Parallel Queues},
  author = {Kamil Demirberk Ünlü and Ali Devin Sezer},
  journal= {arXiv preprint arXiv:1806.00686},
  year   = {2018}
}

Comments

30 pages, 7 figures

R2 v1 2026-06-23T02:17:03.862Z