Abstract
In this paper, we study optimal control problems for multiclass \(GI/M/n+M\) queues in an alternating renewal (up–down) random environment in the Halfin–Whitt regime. Assuming that the downtimes are asymptotically negligible and only the service processes are affected, we show that the limits of the diffusion-scaled state processes under non-anticipative, preemptive, work-conserving scheduling policies, are controlled jump diffusions driven by a compound Poisson jump process. We establish the asymptotic optimality of the infinite-horizon discounted and long-run average (ergodic) problems for the queueing dynamics. Since the process counting the number of customers in each class is not Markov, the usual martingale arguments for convergence of mean empirical measures cannot be applied. We surmount this obstacle by demonstrating the convergence of the generators of an augmented Markovian model which incorporates the age processes of the renewal interarrival times and downtimes. We also establish long-run average moment bounds of the diffusion-scaled queueing processes under some (modified) priority scheduling policies. This is accomplished via Foster–Lyapunov equations for the augmented Markovian model.
Similar content being viewed by others
References
Atar, R., Mandelbaum, A., Reiman, M.I.: Scheduling a multi class queue with many exponential servers: asymptotic optimality in heavy traffic. Ann. Appl. Probab. 14(3), 1084–1134 (2004)
Atar, R.: Scheduling control for queueing systems with many servers: asymptotic optimality in heavy traffic. Ann. Appl. Probab. 15(4), 2606–2650 (2005). https://doi.org/10.1214/105051605000000601
Atar, R., Mandelbaum, A., Shaikhet, G.: Simplified control problems for multiclass many-server queueing systems. Math. Oper. Res. 34(4), 795–812 (2009). https://doi.org/10.1287/moor.1090.0404
Arapostathis, A., Biswas, A., Pang, G.: Ergodic control of multi-class \(M/M/N+M\) queues in the Halfin-Whitt regime. Ann. Appl. Probab. 25(6), 3511–3570 (2015)
Arapostathis, A., Pang, G.: Infinite-horizon average optimality of the N-network in the Halfin-Whitt regime. Math. Oper. Res. 43(3), 838–866 (2018)
Arapostathis, A., Pang, G.: Infinite horizon asymptotic average optimality for large-scale parallel server networks. Stoch. Process. Appl. 129(1), 283–322 (2019)
Budhiraja, A., Ghosh, A., Liu, X.: Scheduling control for Markov-modulated single-server multiclass queueing systems in heavy traffic. Queueing Syst. 78(1), 57–97 (2014)
Kumar, R., Lewis, M.E., Topaloglu, H.: Dynamic service rate control for a single-server queue with Markov-modulated arrivals. Naval Res. Logist. 60(8), 661–677 (2013)
Xia, L., He, Q., Alfa, A.S.: Optimal control of state-dependent service rates in a MAP/M/1 queue. IEEE Trans. Autom. Control 62(10), 4965–4979 (2017)
Arapostathis, A., Das, A., Pang, G., Zheng, Y.: Optimal control of Markov-modulated multiclass many-server queues. Stoch. Syst. 9(2), 155–181 (2019)
Jansen, H. M., Mandjes, M., De Turck, K., Wittevrongel, S.: Diffusion limits for networks of Markov-modulated infinite-server queues. Perform. Eval. 135 (2019). https://doi.org/10.1016/j.peva.2019.102039.
Pang, G., Whitt, W.: Service interruptions in large-scale service systems. Manag. Sci. 55(9), 1499–1512 (2009)
Pang, G., Whitt, W.: Heavy-traffic limits for many-server queues with service interruptions. Queueing Syst. 61(2–3), 167–202 (2009). https://doi.org/10.1007/s11134-009-9104-2
Lu, H., Pang, G., Zhou, Y.: \(G/GI/N(+GI)\) queues with service interruptions in the Halfin-Whitt regime. Math. Methods Oper. Res. 83(1), 127–160 (2016). https://doi.org/10.1007/s00186-015-0523-z
Lu, H., Pang, G.: Heavy-traffic limits for an infinite-server fork-join queueing system with dependent and disruptive services. Queueing Syst. 85(1–2), 67–115 (2017). https://doi.org/10.1007/s11134-016-9505-y
Pang, G., Zhou, Y.: \(G/G/\infty \) queues with renewal alternating interruptions. Adv. Appl. Probab. 48(3), 812–831 (2016). https://doi.org/10.1017/apr.2016.29
Arapostathis, A., Pang, G., Zheng, Y.: Ergodic control of diffusions with compound Poisson jumps under a general structural hypothesis. ArXiv e-prints. arXiv:1908.01068 (2019)
Atar, R., Giat, C., Shimkin, N.: On the asymptotic optimality of the \(c\mu /\theta \) rule under ergodic cost. Queueing Syst. 67(2), 127–144 (2011). https://doi.org/10.1007/s11134-010-9206-x
Konstantopoulos, T., Last, G.: On the use of Lyapunov function methods in renewal theory. Stoch. Process. Appl. 79(1), 165–178 (1999)
Meyn Sean, P., Tweedie, R.L.: Stability of Markovian processes. III. Foster–Lyapunov criteria for continuous-time processes. Adv. Appl. Probab. 25(3), 518–548 (1993)
Billingsley, P.: Convergence of Probability Measures. Wiley Series in Probability and Statistics, Second edn. Wiley, New York (1999). https://doi.org/10.1002/9780470316962. A Wiley-Interscience Publication
Whitt, W.: Stochastic-Process Limits. An Introduction to Stochastic-Process Limits and Their Application to Queues. Springer Series in Operations Research. Springer, New York (2002)
Dai, J.G.: On positive Harris recurrence of multiclass queueing networks: a unified approach via fluid limit models. Ann. Appl. Probab. 5(1), 49–77 (1995)
Arapostathis, A., Pang, G., Sandrić, N.: Ergodicity of a Lévy-driven SDE arising from multiclass many-server queues. Ann. Appl. Probab. 29(2), 1070–1126 (2019)
Arapostathis, A., Hmedi, H., Pang, G., Sandrić, N.: Uniform polynomial rates of convergence for a class of Lévy-driven controlled SDEs arising in multiclass many-server queues. In: Yin, G., Zhang, Q. (eds.) Modeling, Stochastic Control, Optimization, and Applications. The IMA Volumes in Mathematics and its Applications, vol. 164. Springer, Cham (2019)
Davis, M.H.A.: Piecewise-deterministic Markov processes: a general class of nondiffusion stochastic models. J. R. Stat. Soc. Ser. B 46(3), 353–388 (1984). With discussion
Ross, Sheldon M.: Stochastic Processes, 2nd edn. Wiley, New York (1996)
Arapostathis, A., Caffarelli, L., Pang, G., Zheng, Y.: Ergodic control of a class of jump diffusions with finite Lévy measures and rough kernels. SIAM J. Control Optim. 57(2), 1516–1540 (2019)
Krichagina, E.V., Taksar, M.I.: Diffusion approximation for \(GI/G/1\) controlled queues. Queueing Syst. Theory Appl. 12(3–4), 333–367 (1992). https://doi.org/10.1007/BF01158808
Pang, G., Talreja, R., Whitt, W.: Martingale proofs of many-server heavy-traffic limits for Markovian queues. Probab. Surv. 4, 193–267 (2007). https://doi.org/10.1214/06-PS091
Iglehart, D.L., Whitt, W.: The equivalence of functional central limit theorems for counting processes and associated partial sums. Ann. Math. Stat. 42, 1372–1378 (1971). https://doi.org/10.1214/aoms/1177693249
Acknowledgements
This research was supported in part by the Army Research Office through Grant W911NF-17-1-001, in part by the National Science Foundation through Grants DMS-1715210, CMMI-1635410 and DMS-1715875, and in part by the Office of Naval Research through Grant N00014-16-1-2956 and was approved for public release under DCN #43-5442-19.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix A. Proofs of Lemma 3.1 and Proposition 3.1
Proof of Lemma 3.1
By [13, Lemma 5.1], \({\hat{S}}^n_i(t)\) and \({\hat{R}}^n_i(t)\) in (3.1) are martingales with respect to the filtration \({{\mathcal {F}}}^n_t\) in (2.9), having predictable quadratic variation processes given by
respectively. By (2.7), we have the crude inequality
Using the balance equation in (2.5), we see that the same inequalities hold for \(n^{-1}{Z}_i^n\) and \(n^{-1}{Q}_i^n\). Since \(\Psi ^n(s) \in \{0,1\}\), it follows by Lemma 5.8 in [30] that \(\{{\hat{W}}^n_i:n\in \mathbb {N}\}\) is stochastically bounded in \((\mathbb {D}^d,J_1)\). Also, \(\{{\hat{L}}^n_i:n\in \mathbb {N}\}\) is stochastically bounded in \((\mathbb {D}^d,M_1)\) by (2.4). On the other hand, it is evident that
where C is some positive constant. Thus, we obtain
Since \({\hat{X}}^n(0)\) is uniformly bounded, applying Lemma 5.3 in [30] and Gronwall’s inequality, we deduce that \(\{{\hat{X}}^n:n\in \mathbb {N}\}\) is stochastically bounded in \((\mathbb {D}^d,M_1)\). Using Lemma 5.9 in [30], we see that
which implies that \(n^{-1}{X}^n \Rightarrow {\mathfrak {e}}_\rho \) in \((\mathbb {D}^d,M_1)\). By (2.5), and the fact \(\langle e,n^{-1}{Q}^n \rangle = (\langle e,n^{-1}{X}^n\rangle - 1)^ + \Rightarrow {\mathfrak {e}}_0\), we have \(n^{-1}{Q}^n \Rightarrow {\mathfrak {e}}_0\), and thus \(n^{-1}{Z}^n \Rightarrow {\mathfrak {e}}_\rho \). This completes the proof. \(\square \)
To prove Proposition 3.1, we first consider a modified process. Let \(\check{X}^n = (\check{X}^n_1,\dotsc ,\check{X}^n_d)'\) be the d-dimensional process defined by
Lemma A.1
As \(n\rightarrow \infty \), \(\check{X}^n\) and \({\hat{X}}^n\) are asymptotically equivalent, that is, if either of them converges in distribution as \(n\rightarrow \infty \), then so does the other, and both of them have the same limit.
Proof
Let \(K = K(\epsilon _1)>0\) be the constant satisfying \({{\,\mathrm{{\mathbb {P}}}\,}}(\Vert {\hat{X}}^n\Vert _T > K) < \epsilon _1\) for \(T>0\) and any \(\epsilon _1 > 0\), where \(\Vert {\hat{X}}^n\Vert _T :=\sup _{0\le t\le T}\Vert {\hat{X}}^n(t)\Vert \). Since \({\hat{U}}^n(s)\in {{\mathcal {S}}}\) for \(s\ge 0\), on the event \(\{\Vert {\hat{X}}^n\Vert _T \le K\}\), we obtain
where \(C_1\) and \(C_2\) are some positive constants. Then, by Gronwall’s inequality, on the event \(\{\Vert {\hat{X}}^n\Vert _T \le K\}\), we have
Thus, applying [13, Lemma 2.2], we deduce that for any \(\epsilon _2 > 0\), there exist \(\epsilon _3>0\) and \(n_{\circ } = n_{\circ }(\epsilon _1,\epsilon _2,\epsilon _3,T)\) such that
on the event \(\{\Vert {\hat{X}}^n\Vert _T \le K\}\cap \{\Vert C^n_{{\mathsf {d}}}\Vert _T\le \epsilon _3\}\), for all \(n\ge n_{\circ }\), which implies that
As a consequence, \(\Vert \check{X}^n-{\hat{X}}^n\Vert _T\Rightarrow 0\), as \(n\rightarrow \infty \), and this completes the proof. \(\square \)
Proof of Proposition 3.1
We first prove (i). Define the processes
\({\tilde{S}}_i^n(t) :=n^{-\nicefrac {1}{2}}(S^n(nt) - nt)\), and \({\tilde{R}}_i^n(t) :=n^{-\nicefrac {1}{2}}(R^n(nt) - nt)\), for \(i\in {{\mathscr {I}}}\). Then, since \(\Psi ^n(s)\in \{0,1\}\) for \(s\ge 0\), applying Lemma 3.1 and Lemma 2.2 in [13], we have
in \((\mathbb {D}, M_1)\), as \(n\rightarrow \infty \), and that \(\uptau ^n_{2,i}\) weakly converges to the zero process. Since \(\{A_i^n,S_i^n,R^n_i,\Psi ^n:i\in {{\mathscr {I}}}, n\in \mathbb {N}\}\) are independent processes, and \(\tau ^n_{1,i}\) and \(\tau ^n_{2,i}\) converge to deterministic functions, we have joint weak convergence of \(({\hat{A}}^n,{\hat{S}}^n,{\hat{R}}^n,{\hat{L}}^n,\uptau ^n_{1},\uptau ^n_2)\), where \(\uptau ^n_1:=(\uptau ^n_{1,1},\dotsc ,\uptau ^n_{1,d})'\), and \(\uptau ^n_2\) is defined analogously. On the other hand, since the second moment of \(A^n\) is finite, it follows that \({\hat{A}}^n\) converges weakly to a d-dimensional Wiener process with mean 0 and covariance matrix \(\mathrm{diag}\bigl (\sqrt{\lambda _1c^2_{a,1}},\dotsc ,\sqrt{\lambda _dc^2_{a,d}}\bigr )\) (see, e.g., [31]). Therefore, by the FCLT for the Poisson processes \({\tilde{S}}^n\) and \({\tilde{R}}^n\), and using the random time change lemma in [21, Page 151], we obtain (i).
Using (A.1) and Proposition 3.1 (i), the proof of (ii) is same as the proof of [1, Lemma 4 (iii)].
To prove (iii), we first show any limit of \(\check{X}^n\) in (A.2) satisfies (3.3). Following an argument similar to the proof of Lemma 5.2 in [13], one can easily show that the d-dimensional integral mapping \(x = \Lambda (y,u) :\mathbb {D}^d\times \mathbb {D}^d \rightarrow \mathbb {D}^d\) defined by
is continuous in \((\mathbb {D}^d,M_1)\), provided that the function \(h:{\mathbb {R}^{d}}\times {\mathbb {R}^{d}}\rightarrow {\mathbb {R}^{d}}\) is Lipschitz continuous in each coordinate. Since
then, by the tightness of \({U}^n\) and the continuous mapping theorem, any limit of \(\check{X}^n\) satisfies (A.2), and the same result holds for \({\hat{X}}^n\) by Lemma A.1.
Recall the definition of \(\breve{\tau }^n\) in (2.8). It is evident that
for all \(t,r\ge 0\) and \(i\in {{\mathscr {I}}}\). By Assumption 2.2, we have \(\breve{\tau }^n(t) \Rightarrow t\) as \(n\rightarrow \infty \), for \(t\ge 0\). Then, by the random time change lemma in [21, Page 151], we deduce that the last four terms on the right-hand side of (A.3) converge to 0 in distribution. It follows by Proposition 3.1 (i) and (A.3) that
Repeating the same argument we establish convergence of \({\hat{S}}^n\) and \({\hat{R}}^n\). Proving that U is non-anticipative follows exactly as in [1]*Lemma 6. This completes the proof of (iii). \(\square \)
B Proofs of Lemmas 4.1 and 5.2
In this section, we construct two functions, which are used to show the ergodicity of \({\widetilde{\Xi }}^n\). We provide two lemmas concerning the properties of these functions, respectively. The proofs of Lemmas 4.1 and 5.2 are given at the end of this section.
Definition B.1
For \(z^n \in {\mathfrak {Z}}_{\mathrm {sm}}^n\), define the operator \({\mathcal {L}}^{z^n}_n:{\mathcal {C}}_b({\mathbb {R}^{d}}\times {\mathbb {R}^{d}}) \rightarrow {\mathcal {C}}_b({\mathbb {R}^{d}}\times {\mathbb {R}^{d}})\) by
for \(f\in {\mathcal {C}}_b({\mathbb {R}^{d}}\times {\mathbb {R}^{d}})\) and any \((\breve{x},h)\in \mathbb {R}^d_+\times \mathbb {R}^d_+\), with \(q^n :=\breve{x} - z^n\).
Note that if \(d_1^n \equiv 0\) for all n, the queueing system has no interruptions. In this situation, under a Markov scheduling policy, the (infinitesimal) generator of \((X^n,H^n)\) takes the form of (B.1). Recall the scheduling policies \(\check{z}^n\) in Definition 4.1, and \({\bar{x}} = \breve{x} - n\rho \) in Definition 4.2. We define the sets
We have the following lemma.
Lemma B.1
Grant Assumptions 2.1, 2.2, and 3.2. For any even integer \(\upkappa \ge 2\), there exist a positive vector \(\xi \in \mathbb {R}^d_+\), \(\breve{n}\in \mathbb {N}\), and positive constants \(\breve{C}_0\) and \(\breve{C}_1\), such that the functions \(f_n\), \(n\in \mathbb {N}\), defined by
with \(\eta ^n_i\) as defined in (4.3), satisfy
for all \(n\ge \breve{n}\) and \((\breve{x},h)\in \mathbb {R}^d_+\times \mathbb {R}^d_+\).
Proof
Using the estimate
an easy calculation shows that
where for the fourth term on the right-hand side we also used the fact that
It is clear that \(\eta ^n_i(0) = 0\), since \(F_i(0) = 0\) and \({{\,\mathrm{{\mathbb {E}}}\,}}[G_i] = 1\). On the other hand, \(\eta ^n_i(t)\) is bounded for all \(n\in \mathbb {N}\) and \(t\ge 0\) by Assumption 3.2. Thus, applying (B.4), (B.5) and (4.4), it follows that
Since \(\eta ^n_i(h_i)\) is uniformly bounded, and \(\check{z}^n_i, \check{q}^n_i \le {\bar{x}}_i + n\rho _i\), it follows that the last term in (B.6) is equal to \({{\mathscr {O}}}(n){{\mathscr {O}}}(|{\bar{x}}_i|^{\upkappa -2}) + {{\mathscr {O}}}(|{\bar{x}}_i|^{\upkappa -1})\). Note that for \(i\in {{\mathscr {I}}}\setminus {{\mathscr {I}}}_0\), \(\check{z}^n_i\) is equivalent to the static priority scheduling policy. Note also, that
and for \(i\in {{\mathscr {I}}}_0 \setminus {\tilde{{{\mathcal {K}}}}}_n(\breve{x})\), we have \(\check{z}^n_i - n\rho _i = {\bar{x}}_i\) and \(\check{q}^n_i = 0\). By using (B.6), and the identity in (5.20), we obtain
Let \(\breve{c}_1:=\sup _{i,n}\{\gamma ^n_i,\mu ^n_i\}\), and \(\breve{c}_2\) be some constant such that \(\inf \{\mu _i^n,\gamma ^n_j :i\in {{\mathscr {I}}},j\in {{\mathscr {I}}}\setminus {{\mathscr {I}}}_0, n \in \mathbb {N}\} \ge \breve{c}_2 > 0\). We select a positive vector \(\xi \in \mathbb {R}^d_+\) such that \(\xi _1 :=1\), \(\xi _{i}:=\frac{\kappa ^m_1}{d^\upkappa } \min _{i^{\prime }\le i-1}\xi _{i^{\prime }}\), \(i\ge 2\), with \(\kappa _1 :=\frac{\breve{c}_1}{8\breve{c}_2}\). Compared with [4, Lemma 5.1], the important difference here is that, for \(i\in {{\mathscr {I}}}\setminus {{\mathscr {I}}}_0\), we have
Repeating the argument in the proof of [4, Lemma 5.1], it follows by (B.8) that
for some positive constants \(c_3\), \(c_4\) and \(c_5\). Therefore, (B.3) follows by (B.7) and (B.9), and this completes the proof. \(\square \)
Let
for \((\breve{x},h,\psi ,k)\in {{\mathfrak {D}}}\), where \({\tilde{g}}_{n,i}(\breve{x}_i) :=- |{\bar{x}}_i|^{\upkappa }\) for \(i\in {{\mathscr {I}}}\setminus {{\mathscr {I}}}_0\), and
Recall \({\overline{{\mathcal {L}}}}^{z^n}_{n,\psi }\) in (4.13). We also define
Lemma B.2
Grant Assumptions 2.1, 2.2 and 3.2, and let \(\xi \in \mathbb {R}^d_+\) be as in (B.2). Then, for any even integer \(\upkappa \ge 2\) and any \(\varepsilon >0\), there exist a positive constant \({\overline{C}}\), and \({\bar{n}}\in \mathbb {N}\), such that
for any \(z^n\in {\mathfrak {Z}}_{\mathrm {sm}}^n\), and all \((\breve{x},h,\psi ,k)\in {{\mathfrak {D}}}\) and \(n>{\bar{n}}\).
Proof
It is straightforward to verify that
for \(i\in {{\mathscr {I}}}\). Repeating the calculation in (B.5) and (B.6), and applying (B.4) and (B.12), we have
Note that \({\overline{q}}^{n,k}_i \le c(1 + \langle e,{\bar{x}}\rangle ^{+})\) for some positive constant c, by (5.21). Since \({z}^n_i, {q}^n_i \le {\bar{x}}_i + n\rho _i\), \((\vartheta ^n)^{-1}\) is of order \(n^{-\nicefrac {1}{2}}\) by Assumption 2.2, and \(\eta ^n_i\) and \(\upalpha ^n\) are bounded, it follows by (5.20) and (B.13) that
Thus, applying Young’s inequality, we obtain (B.11), and this completes the proof. \(\square \)
Proof of Lemma 4.1
We define the function \({\tilde{f}}_n \in {\mathcal {C}}({\mathbb {R}^{d}}\times \mathbb {R}^d_+\times \{0,1\}\times \mathbb {R}_+)\) by
with \(f_n\) and \({\tilde{g}}_n\) in (B.2) and (B.10), respectively. Recall \({\widetilde{{{\mathcal {V}}}}}^n_{\upkappa ,\xi }\) in (4.17). With \(\xi \in \mathbb {R}^d_+\) as in (B.2), we have
Hence, to prove (4.18), it suffices to show that
and all \((\breve{x},h,\psi ,k)\in {{\mathfrak {D}}}\), where the generator \(\breve{{\mathcal {L}}}^{\check{z}^n}_n\) is given in (4.12). It is clear that \({{\mathcal {Q}}}_{n,\psi }f_n(\breve{x},h) = 0\). Since \((\vartheta ^n)^{-1}\) is of order \(n^{\nicefrac {-1}{2}}\), it follows by (4.10) and (4.15) that
where C is some positive constant and \(\epsilon _n\rightarrow 0\) as \(n\rightarrow \infty \). Since all the moments of \(d_1\) are finite by (3.10) and \((a+z)^\upkappa - a^\upkappa = {{\mathscr {O}}}(z){{\mathscr {O}}}(a^{\upkappa -1}) + {{\mathscr {O}}}(z^2){{\mathscr {O}}}(a^{\upkappa -2})+ \dots + {{\mathscr {O}}}(z^\upkappa )\) for any \(a,z\in \mathbb {R}\), it is easy to verify that
using also the fact that
which follows by by Assumptions 2.1, 2.2, and (3.10). Then, for \(\psi =1\), it follows by (B.16) and Young’s inequality that
Note that the last two terms in (B.3) and the last term in (B.11) are of smaller order than the second and third terms on the right-hand side of (B.3), respectively. Thus, applying Lemmas B.1 and B.2, and using (B.17), we obtain
for all large enough n, where \({\tilde{x}}\) is defined in Definition 4.2. Since \(\check{q}^n_i \ge 0\) and \(\check{z}^n_i - n\rho _i>0\) for \(i\in {{\tilde{{{\mathcal {K}}}}}}_n(\breve{x})\), then by using (B.7) and (B.18), we see that (B.14) holds when \(y = 1\).
For \(\psi = 0\), using (B.15), Young’s inequality, and the fact that for \(i\in {\tilde{{{\mathcal {K}}}}}_n(\breve{x})\), \({\bar{x}}_i > 0\), we obtain
for some positive constant C and sufficiently small \(\epsilon >0\). We proceed by invoking the argument in the proof of [4, Lemma 5.1]. The important difference here is that
where the functions \({\tilde{\epsilon }}_i, {\bar{\epsilon }}_i :{\mathbb {R}^{d}}\rightarrow [0,1]\), for \(i\in {{\mathscr {I}}}\). Since \({\tilde{\epsilon }}_i\) and \({\bar{\epsilon }}_i\) are bounded, we have some additional terms which are bounded by \(C \int _{\mathbb {R}_*}n\mu _i\rho _i(y - k)\,{\tilde{F}}^{d^n_1}_{\breve{x},k}(\mathrm {d}{y})\) for some positive constant C. Therefore, these are of order \(\sqrt{n}\) by (5.21). Thus, repeating the argument in the proof of Lemma B.1, and applying Lemma B.2, we deduce that (B.14) holds with \(\psi =0\). This completes the proof. \(\square \)
Proof of Lemma 5.2
The proof mimics that of Proposition 4.2. We sketch the proof when \({{\mathscr {I}}}_0\) is empty. Using the estimate
for any \(\epsilon >0\), which follows by Young’s inequality, we deduce that, for some positive constants \(\{c_k :k=1,2,3\}\), we have
and all large enough n. Note that Lemma B.2 holds for all \(z^n\in {\mathfrak {Z}}_{\mathrm {sm}}^n\). Then, we may repeat the steps in the proof of Lemma 4.1, except that here we use
where \({\hat{q}}^n = n^{\nicefrac {-1}{2}}q^n\), with \(\epsilon >0\) chosen sufficiently small. Since \({\hat{q}}^n_i(\breve{x},z^n) \le \langle e,{\tilde{x}} \rangle ^+\), it follows by (5.21) that
Thus, by the same calculation in Proposition 4.2, and using (B.19)–(B.22), we obtain
for all large enough n, and \(\{z^n\in {\mathfrak {Z}}_{\mathrm {sm}}^n:n\in \mathbb {N}\}\). Since \(\sup _n {\hat{J}}({\hat{X}}^n(0),{z}^n)<\infty \), it follows by (4.26) that
Therefore, dividing both sides of (B.23) by T, taking \(T\rightarrow \infty \) and using (4.26) again, we obtain (5.11). We may show that the result also holds when \({{\mathscr {I}}}_0\) is nonempty by repeating the above argument and applying Lemma B.2. This completes the proof. \(\square \)
Rights and permissions
About this article
Cite this article
Arapostathis, A., Pang, G. & Zheng, Y. Optimal Scheduling of Critically Loaded Multiclass GI/M/n+M Queues in an Alternating Renewal Environment. Appl Math Optim 84, 1857–1901 (2021). https://doi.org/10.1007/s00245-020-09698-9
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00245-020-09698-9
Keywords
- Multiclass many-server queues
- Halfin–Whitt (QED) regime
- Service interruptions
- Renewal arrivals
- Alternating renewal process
- Jump diffusions
- Discounted cost
- Ergodic control
- Asymptotic optimality