Abstract
We consider a state-dependent single-server queue with orbit. This is a versatile model for the study of service systems, where the server needs a non-negligible time to retrieve waiting customers every time he completes a service. This situation arises typically when the customers are not physically present at a system, but they have a remote access to it, as in a call center station, a communication node, etc. We introduce a probabilistic approach for the performance evaluation of this queueing system, that we refer to as the queueing and Markov chain decomposition approach. Moreover, we discuss the applicability of this approach for the performance evaluation of other non-Markovian service systems with state dependencies.
Similar content being viewed by others
References
Abouee-Mehrizi, H., Baron, O.: State-dependent \(M/G/1\) queueing systems. Queueing Syst. 82, 121–148 (2016)
Abouee-Mehrizi, H., Baron, O., Berman, O.: Exact analysis of capacitated two-echelon inventory systems with priorities. Manuf. Serv. Oper. Manag. 16, 561–577 (2014)
Artalejo, J.R., Gomez-Corral, A.: Retrial Queueing Systems. A Computational Approach. Springer, Berlin (2008)
Baron, O., Economou, A., Manou, A.: Equilibrium customer strategies for joining the \(M/G/1\) queue with orbit. Working paper (2017)
Bitran, G.R., Tirupati, D.: Multiproduct queueing networks with deterministic routing: decomposition approach and the notion of interference. Manage. Sci. 34, 75–100 (1988)
Caldentey, R.: Approximations for multi-class departure processes. Queueing Syst. 38, 205–212 (2001)
Choi, B.D., Park, K.K., Pearce, C.E.M.: An \(M/M/1\) retrial queue with control policy and general retrial times. Queueing Syst. 14, 275–292 (1993)
Cox, D.R.: A use of complex probabilities in the theory of stochastic processes. Proc. Camb. Philos. Soc. 51, 313–319 (1955)
Doroudi, S., Fralix, B., Harchol-Balter, M.: Clearing analysis of phases: exact limiting probabilities for skip-free, unidirectional, quasi-birth–death processes. Stoch. Syst. 6, 420–458 (2016)
Eager, D.L., Lazowska, E.D., Zahorjan, J.: Adaptive load sharing in homogeneous distributed systems. IEEE Trans. Softw. Eng. 12, 662–675 (1986)
Economou, A., Manou, A. (2015) A probabilistic approach for the analysis of the \(M_n/G/1\) queue. Ann. Oper. Res. https://doi.org/10.1007/s10479-015-1943-0
Falin, G.I., Templeton, J.G.C.: Retrial Queues. Chapman and Hall, London (1997)
Fayolle, G.: A simple telephone exchange with delayed feedbacks. In: Boxma, O.J., Cohen, J.W., Tijms, H.C. (eds.) Teletraffic Analysis and Computer Performance Evaluation, pp. 245–253. Elsevier, Amsterdam (1986)
Gomez-Corral, A.: Stochastic analysis of a single server retrial queue with general retrial times. Naval Res. Logist. 46, 561–581 (1999)
Kerner, Y.: The conditional distribution of the residual service time in the \(M_n/G/1\) queue. Stoch. Models 24, 364–375 (2008)
Lazowska, E.D., Zahorjan, J., Cheriton, D.R., Zwaencpocl, W.: File access performance of diskless workstations. Department of Computer Science, University of Washington, Seattle, Technical Report 84-06-01 (1984)
Manou, A., Economou, A., Karaesmen, F.: Strategic customers in a transportation station: when is it optimal to wait? Oper. Res. 62, 910–925 (2014)
Martin, M., Artalejo, J.R.: Analysis of an \(M/G/1\) queue with two types of impatient units. Adv. Appl. Prob. 27, 840–861 (1995)
Oz, B., Adan, I., Haviv, M.: A rate balance principle and its application to queueing models. Queueing Syst. 87, 95–111 (2017)
Oz, B., Adan, I., Haviv, M.: The conditional distribution of the remaining service or vacation time in the \(M_n/G_n/1\) queue with vacations. Preprint (2015)
Oz, B.: Strategic behavior in queues. Ph.D. Thesis, The Hebrew University of Jerusalem (2016)
Phung-Duc, T., Rogiest, W., Wittevrongel, S.: Single server retrial queues with speed scaling: analysis and performance evaluation. J. Ind. Manag. Optim. (2017). https://doi.org/10.3934/jimo.2017025
Phung-Duc, T.: Server farms with batch arrival and staggered setup. Proceedings of the 15th Symposium on Information and Communication Technology (SoICT), Hanoi, December 4–5, 2014, pp. 240–247 (2014)
Phung-Duc, T.: Exact solutions for \(M/M/c/\text{ Setup }\) queues. Telecommun. Syst. 64, 309–324 (2017a)
Phung-Duc, T.: Single server retrial queues with setup time. J. Ind. Manag. Optim. 13, 1329–1345 (2017b)
Reiser, M., Kobayashi, H.: Accuracy of diffusion approximation for some queueing systems. IBM J. Res. Dev. 18, 110–124 (1974)
van Doorn, E.A., Regterschot, D.J.K.: Conditional PASTA. Oper. Res. Lett. 7, 229–232 (1988)
Wang, J., Baron, O., Scheller-Wolf, A.: \(M/M/c\) queue with two priority classes. Oper. Res. 63, 733–749 (2015)
Yajima, M., Phung-Duc, T.: Batch arrival single-server queue with variable service speed and setup time. Queueing Syst. 86, 241–260 (2017)
Acknowledgements
We thank the associate editor, and two anonymous reviewers for their valuable comments that greatly improved the paper. We thank Professor Mor Harchol-Balter of Carnegie Mellon University for pointing us to the application of our model for the analysis of load sharing algorithms. Professor Baron’s work on this research was supported by a grant from the Natural Science and Engineering Research Council of Canada. Athanasia Manou was supported by AXA Research Fund.
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
In this Appendix, we first provide several technical analytic proofs of various results of the paper. We then use the supplementary variable method to prove the main results in the paper. The proofs are quite tedious but straightforward, depending mainly on algebraic manipulations, and they are given for completeness.
1.1 Technical analytic proofs
Corollary 5.2—Proof
In light of (5.17), (5.19) yields
Now, by substituting \(\mu _{1j}\) from the RHS of (5.18) into (5.20), canceling \(\lambda _{0j} p_{0j}\) yields \( \lambda _{1,j-1}p_{1,j-1}+\mu _{0,j+1}=\mu _{0j}+\lambda _{1j} p_{1j}\), for \( j\ge 1\), which can be written as
Summing (7.2) for consecutive values of j and canceling equal terms yields
which reduces to \(\mu _{0,j+1}=\lambda _{1j}p_{1j}\), for \(j\ge 1\), using (7.1). \(\square \)
Lemma 5.2—Proof
We have
\(\square \)
Theorem 5.1—Proof
We first establish the initial conditions. The initial condition (5.40) is simply (5.21). By (5.40) and (5.34) we have that
which yields (5.41). By (5.33) for \(j=1\) we have
which, in combination with (5.41), yields (5.42). Also, (5.43) is identical to (5.8).
Now, solving (5.35) for \(\mu _{0,j+1}\) yields (5.36). Indeed, we have that
where in the second equality we used equations (5.22) and (5.23). Now, (5.37) immediately follows from (5.33) for \(j\ge 1\) and (5.38) is identical to (5.9). Substituting (5.1) into (5.13) and using (5.22) yields
which gives the first line in (5.39). The second line follows using (5.1), (5.2) and (5.22). Indeed, we have
Finally, (5.44) is identical to (5.7).\(\square \)
Theorem 5.2—Proof
Using (5.22) and (5.23), we deduce that
Equations (5.23) and (5.33) yield
so, using (7.3), we conclude that
Solving for \(p_{0j}\) yields (5.45). Next, equations (5.34) and (5.21) imply that
where the second and third equality follow by (5.2) and (5.24), respectively. Noting that \(\mu _{01}=\lambda _{10}p_{10}\) from (5.23) and solving for \(p_{10}\) yields both expressions in (5.47).
Finally, substituting (7.3) and (5.23) into the left-hand side and right-hand side of (5.35), respectively, we get
Rearranging terms in the first equality yields
Solving for \(p_{1j}\) yields the first line in (5.46). Substituting (5.24) in the second line of (7.4) implies that
Solving for \(p_{1j}\) yields the second line in (5.46). \(\square \)
Corollary 5.4—Proof
We prove the result by induction. First, note that (5.40)–(5.42) imply that \(\mu _{10}, \mu _{01}\) and \(\mu _{11}\) are linear in \(p_{00}\). Also, (5.44) shows that LSTs \(\tilde{S}_{0j}(s)\) are independent of \(p_{00}\), for \(j\ge 1\). Since \(\mu _{01}\) is linear in \(p_{00}\), (5.1) for \(j=1\) implies that \(\nu _{10}\) is linear in \(p_{00}\). Moreover, since \(\nu _{10}\) and \(\mu _{01}\) are linear in \(p_{00}\), (5.2) implies that \(\tilde{B}_{0}(s)\) is independent of \(p_{00}\) and consequently \(\tilde{S}_{10}(s)\) is independent of \(p_{00}\). From formulas (5.45) for \(j=1\) and (5.47), we have that \(p_{01}\) and \(p_{10}\) are linear in \(p_{00}\).
We make the induction hypothesis: Assume that the rates \(\mu _{01}, \mu _{02},\ldots , \mu _{0j}\), the rates \(\mu _{10}, \mu _{11},\ldots , \mu _{1j}\), the rates \(\nu _{10}, \nu _{11},\ldots , \nu _{1,j-1}\), the probabilities \(p_{10}, p_{11},\ldots , p_{1,j-1}\), and the probabilities \(p_{01}, p_{02},\ldots , p_{0j}\) are linear in \(p_{00}\). Moreover, assume that the LSTs \(\tilde{B}_{0}(s), \tilde{B}_{1}(s),\ldots , \tilde{B}_{j-1}(s)\) and \(\tilde{S}_{10}(s), \tilde{S}_{11}(s),\ldots , \tilde{S}_{1,j-1}(s)\) are independent of \(p_{00}\).
We will prove the following: (i) \(\mu _{0,j+1}\) is linear in \(p_{00}\), (ii) \(\mu _{1,j+1}\) is linear in \(p_{00}\), (iii) \(\nu _{1j}\) is linear in \(p_{00}\), (iv) \(\tilde{B}_{j}(s)\) is independent of \(p_{00}\), (v) \(\tilde{S}_{1j}(s)\) is independent of \(p_{00}\), (vi) \(p_{1j}\) is linear in \(p_{00}\) and (vii) \(p_{0,j+1}\) is linear in \(p_{00}\).
-
(i)
Since \(\mu _{0j}\) and \(\mu _{1j}\) are linear in \(p_{00}\) and \(\tilde{S}_{1,j-1}(s)\) is independent of \(p_{00}\), (5.36) implies that \(\mu _{0,j+1}\) is linear in \(p_{00}\).
-
(ii)
Using that \(\mu _{0,j+1}\) is linear in \(p_{00}\), (5.37) gives that \(\mu _{1,j+1}\) is linear in \(p_{00}\).
-
(iii)
Since \(\mu _{0,j+1}\) and \(p_{0j}\) are linear in \(p_{00}\), (5.1) implies that \(\nu _{1j}\) is linear in \(p_{00}\).
-
(iv)
Since \(\mu _{0,j+1}, p_{0j}\) and \(\nu _{1j}\) are linear in \(p_{00}\), (5.2) implies that \(\tilde{B}_{j}(s)\) is independent of \(p_{00}\).
-
(v)
Using that \(\mu _{1j}, \mu _{0j}\), and \(\mu _{0,j+1}\) are linear in \(p_{00}\), (5.39) gives that \(a_j\) is independent of \(p_{00}\). Then, using that \(a_j, \tilde{S}_{1,j-1}(s)\) and \(\tilde{B}_{j}(s)\) are independent of \(p_{00}\), (5.38) implies that \(\tilde{S}_{1j}(s)\) is independent of \(p_{00}\).
-
(vi)
Since \(p_{0j}\) and \(p_{1,j-1}\) are linear in \(p_{00}\) and \(\tilde{S}_{1,j-1}(s)\) is independent of \(p_{00}\), (5.46) shows that \(p_{1j}\) is linear in \(p_{00}\).
-
(vii)
(5.45), using that \(p_{1j}\) is linear in \(p_{00}\), yields that \(p_{0,j+1}\) is linear in \(p_{00}\). \(\square \)
Theorem 5.3—Proof
Using (5.46) in combination with (5.45) gives
Thus,
Also, (5.45) and (5.47) can be written as
and
respectively. Then,
The normalization equation yields
Thus, the stability condition is given by (5.48) and \(p_{00}\) is given by (5.49). \(\square \)
Theorem 5.4—Proof
Let \(\tilde{P}_{ij}(s)=p_{ij}\tilde{S}_{ij}(s)\) be the Laplace transform of the steady-state density
i.e.,
Multiplying (5.7)–(5.9) with the corresponding \(p_{ij}\)s yields the following equations for \(\tilde{P}_{ij}(s)\):
We simplify the coefficients appearing in these equations, using (5.21)–(5.23) and (5.33)–(5.35) of the second and third QMCD system. We have that
so we can easily see (using (5.13), (5.21)–(5.23) and (5.33)–(5.35)) that
and
Substituting (7.12)–(7.18) in (7.9)–(7.11) yields
We now apply a generating function approach to determine the equilibrium distribution of the number of customers in the system. To this end, we introduce the generating functions
Multiplying (7.19) by \((s-\lambda )z^j\), and summing for \(j\ge 1\), yields
Similarly, multiplying (7.21) by \((s-\lambda )z^j\), summing for \(j\ge 1\), and summing also (7.20) yields
Note, however, that
so (7.23) assumes the form
Setting \(s=0\) in (7.22), \(s=0\) in (7.24), \(s=\lambda \) in (7.22), and \(s=\lambda (1-z)\) in (7.24) yields the system
in the unknowns \(\tilde{M}_0(z), \tilde{M}_1(z), \tilde{P}_0(0,z)\), and \(\tilde{P}_1(0,z)\). We can easily solve the system by using sequentially (7.27), (7.25), and (7.26), to express the unknowns \(\tilde{M}_1(z), \tilde{P}_0(0,z)\), and \(\tilde{P}_1(0,z)\) in terms of \(\tilde{M}_0(z)\). We get
We can then plug (7.28)–(7.30) and obtain the equation
for \(\tilde{M}_0(z)\). Solving (7.31) for \(\tilde{M}_0(z)\) and substituting in (7.28)–(7.30) yields the following explicit expressions for \(\tilde{M}_0(z), \tilde{M}_1(z), \tilde{P}_0(0,z)\), and \(\tilde{P}_1(0,z)\):
The generating function of the number of customers in the system in steady-state is
Using (7.32)–(7.33), we obtain
The probability of an empty system \(p_{00}\) is determined from the normalization equation \(K(1)=1\). Using L’Hospital’s rule, we obtain
The system is stable if and only if \(p_{00}>0\), which gives the stability condition (5.50). Substituting (7.35) in (7.34) yields (5.51). \(\square \)
1.2 Alternative proofs using the supplementary variable method
In the rest of the Appendix, we use the well-known supplementary variable method, introduced by Cox [8], to provide analytic proofs for the three QMCD Systems (Corollaries 5.1–5.3) which constitute the basis for the recursive schemes for the performance analysis of the model.
To this end, we introduce the following transient probabilities and densities:
and their steady-state counterparts
A moment of reflection reveals that the analytic quantity \(p_{ij}(0)\) is identical to the corresponding rate \(\mu _{ij}\) of service/seeking time completions that we considered in the probabilistic derivations, for all i, j. Therefore, similarly to (5.1) and (5.2), we have in the present analytic framework the equations
Moreover, we denote by \(\tilde{P}_{ij}(s)\) the Laplace transform (LT) of the steady-state density \(p_{ij}(r)\), i.e.,
We consider the evolution of the continuous time Markov process \( \{(C(t),Q(t),S\left( t\right) ):t\ge 0\}\) in the interval \([t,t+\mathrm{d}t]\). Then, we have Lemma 7.1.
Lemma 7.1
The steady-state probabilities \(p_{0j}, j\ge 1,\) the steady-state densities \(p_{ij}(0)\) and the LTs \(\tilde{P}_{ij}(s), (i,j)\in \{0,1\}\times \{0,1,2,\ldots \}{\setminus }\{(0,0)\}\), satisfy the following system of equations:
Proof
Considering the evolution of \(\{(C(t),Q(t),R(t)):t\ge 0\}\) in the interval \( [0,t+\mathrm{d}t]\) and conditioning on its value at time t, we have the equations
for \(\mathrm{d}t\rightarrow 0^+\). Taking the limits of (7.46)–(7.49) as \(t\rightarrow \infty , \) and using (7.36)–(7.38) yields
for \(\mathrm{d}t\rightarrow 0^+\). Rearranging the terms appropriately, dividing by \(\mathrm{d}t \) and taking the limits as \(\mathrm{d}t\rightarrow 0^+\) in (7.50)–(7.53) yields
Equation (7.54) yields immediately (7.42). Also, multiplying (7.55)–(7.57) by \(\mathrm{e}^{-sr}\) and integrating with respect to \( s\in [0,\infty )\) yields equations (7.43)–(7.45).\(\square \)
Now, we can provide an analytic proof of the second and the third QMCD systems.
Lemma 7.2
The steady-state probabilities \(p_{ij}\) and the steady-state densities \(p_{ij}(0)\) satisfy the second QMCD system:
Together with the LSTs \(\tilde{S}_{ij}(s)\), they also satisfy the third QMCD system:
Proof
Equation (7.42) yields immediately (7.58). Taking \( s=0\) in (7.43)–(7.45) yields
Equation (7.64) is identical to Eq. (7.59). Also, using (7.58), Eq. (7.65) yields (7.60) for \(j=0\).
Using (7.59), Eq. (7.66) assumes the form
Iterating this equation yields
Then, using (7.60) for \(j=0\), we deduce (7.60) for \(j\ge 1\).
Equation (7.43) for \(s=\lambda _{0j}\) yields (7.61). Also, setting \(s=\lambda _{10}\) in Eq. (7.44) and using (7.40) yields (7.62). Similarly, Eq. (7.45) for \( s=\lambda _{1j}\) in combination with (7.40) and (7.41) yields (7.63). \(\square \)
Finally, we provide an analytic proof of the first QMCD system, i.e., of Corollary 5.1 with \(a_j\) given by (5.13).
Corollary 5.1—Proof
Analytic proof of (5.7)–(5.9) with \(a_j\) given by (5.13).
Equation (7.43) can be written as
Using (7.61), the above equation yields
Dividing by \(p_{0j}\), we deduce that
which proves formula (5.7).
Equation (7.44) can be written as
Using (7.62) we obtain
and dividing by \(p_{10}\) yields
which proves formula (5.8).
Similarly, Eq. (7.45) is written as
Using (7.63) and (7.41) we obtain
Dividing by \(p_{1j}\) yields
where
and
Thus, formula (5.9) has been proved with \(a_j\) given by (5.13). \(\square \)
Rights and permissions
About this article
Cite this article
Baron, O., Economou, A. & Manou, A. The state-dependent M / G / 1 queue with orbit. Queueing Syst 90, 89–123 (2018). https://doi.org/10.1007/s11134-018-9582-1
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11134-018-9582-1
Keywords
- State-dependent queueing system
- Orbit
- Retrial queue
- Non-negligible retrieval time
- Steady-state distribution
- Conditional sojourn time distributions
- Variable arrival rate
- Variable service speed
- Queueing and Markov chain decomposition