Abstract
This paper analyzes an unreliable \(M^{X}/G(M/H)/1\) retrial system with vacation. We present closed-form expressions for the important performance indicators of the system, and derive the optimal vacation policies for minimizing the average waiting time of orbiting customers. The performance metrics relevant for helpline services are developed. Numerical experiments are conducted to examine the effect of vacation policy on the queue length and busy period of the system.
Similar content being viewed by others
Notes
MAV policies are based on the number of vacations taken before the first caller arrives. The server takes vacation consecutively until at least one caller is found waiting in the system at the vacation completion instant given a maximum number of vacations allowed. If no caller is found in the system after the end of the last allowable vacation, server goes into idling and waits until one caller arrives. MAV policy is first proposed by Takagi (1991) and later by Ke and Chu (2006) and Ke (2007) who refer it as “modified vacation policy”.
References
Aissani, A. (2000). An \(M^{(x)}/G/1\) retrial queue with exhaustive vacations. Journal of Statistics and Management Systems, 3, 269–286.
Artalejo, J. R. (1997). Analysis of an \(M/G/1\) retrial queue with constant rate of repeated attempts and server vacations. Computers and Operations Research, 24, 493–504.
Artalejo, J. R. (1999a). A classified bibliography of research on retrial queues: progress in 1990–1999. TOP, 7(2), 187–211.
Artalejo, J. R. (1999b). Accessible biblography on retrial queues. Mathematical and Computer Modelling, 30, 1–6.
Artalejo, J. R., Falin, G. I., & Lopez-Herrero, M. J. (2002). A second order analysis of the waiting time in the \(M/G/1\) retrial queue. Asia-Pacific Journal of Operational Research, 19(2), 131–148.
Artalejo, J. R., & Gomez-Corral, A. (2008). Retrial queueing systems: A computational approach. New York: Springer.
Atencia, I., Bouza, G., & Moreno, P. (2008). An \(M^{(x)}/G/1\) retrial queue with server breakdowns and constant rate of repeated attempts. Annals of Operations Research, 157, 225–243.
Chang, F. M., & Ke, J. C. (2009). On a batch retrial model with \(J\) vacation. Journal of Computational and Applied Mathematics, 232, 402–414.
Choi, B. D., Shin, Y. W., & Ahn, W. C. (1992). Retrial queues with collision arising from unslooted CSMA/CD Protocols. Queueing Systems, 11, 335–356.
Falin, G. (1990). A survey of retrial queues. Queueing Systems, 7, 127–168.
Falin, G. (2008). An \(M/M/1\) retrial queue with retrial due to server failures. Queueing Systems, 58(3), 155–160.
Falin, G. (2010). An \(M/G/1\) retrial queue with an unreliable server and general repair times. Performance Evaluation, 67(7), 569–582.
Falin, G., & Templeton, J. G. C. (1997). Retrial queues, monographs on statistics and applied probability. London: Chapman and Hall.
Farahmand, K. (1990). Single line queue with repeated demands. Queueing Systems, 6, 223–228.
Fayolle, G. (1986) A simple telephone exchange with delayed feedbacks. In Teletraffic analysis and computer performance evaluation (pp. 245–253) Amsterdam: North Holland.
Ke, J. C. (2003). Optimal strategy policy in batch arrival queue with server breakdowns and multiple vacations. Mathematical Methods of Operations Research, 58, 41–56.
Ke, J. C. (2006). An \(M/G/1\) queue under hysteretic vacation policy with an early startup and unreliable server. Mathematical Methods of Operations Research, 63, 357–369.
Ke, J. C. (2007). Batch arrival queues under vacation policies with server breakdowns and startup/closedown times. Applied Mathematical Modelling, 31, 1282–1292.
Ke, J. C., & Chu, Y. K. (2006). A modified vacation model \(M^{(x)}/G/1\) system. Applied Stochastic Models in Business and Industry, 22, 1–16.
Krishna Kumar, B., Arivudainambi, D., & Vijayakumar, A. (2002). On the \(M^{(x)}/G/1\) retrial queue with Bernoulli schedules and general retrial times. Asia-Pacific Journal of Operational Research, 19(2), 177–194.
Kulkarni, V. G., & Choi, B. D. (1990). Retrial queues with server subject to breakdowns and repairs. Queueing Systems, 7, 191–208.
Li, Q., Ying, Y., & Zhao, Y. Q. (2006). A \(BMAP/G/1\) retrial queue with a server subject to breakdowns and repairs. Annals of Operations Research, 141, 233–270.
Li, H., & Zhao, Y. Q. (2005). A retrial queue with constant retrial rate, server downs and impatient customers. Stochastic Models, 21, 531–550.
Senott, L. I., Humblet, R. L., & Tweedie, R. L. (1983). Mean drifts and non-ergodicity of Markov chain. Operations Research, 31, 783–789.
Senthilkumar, M., & Arumuganathan, R. (2008). On the single server batch arrival retrial queue with general vacation time under Bernoulli schedule and two phases of heterogeneous service. Quality Technology and Quantitative Management, 5(2), 145–160.
Song, J. S. (1994). The effect of leadtime uncertainty in a simple stochastic inventory model. Management Science, 40(5), 603–613.
Takagi, H. (1991). Queueing analysis: Vacation and priority systems. Amsterdam: North-Holland, Elsevier.
Tang, Y. H. (1997). A single server \(M/G/1\) queueing system subject to breakdowns—some reliability and queueing problem. Microelectronics and Reliability, 37, 315–321.
Tian, N. S., & Zhang, Z. G. (2006). Vacation queueing models. New York: Springer.
Wang, J., Cao, J., & Li, Q. (2001). Reliability analysis of the retrial queue with server breakdowns and repairs. Queueing Systems, 38(4), 363–380.
Yang, T., & Li, H. (1994). The M/G/1 retrial queue with the server subject to starting failures. Queueing Systems, 16, 83–96.
Yang, T., & Templeton, J. G. C. (1987). A survey of retrial queues. Queueing Systems, 2, 201–233.
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
Notations | Meaning |
---|---|
\(\mathcal {P}\) | Vacation policy |
\(M^{X}/G/1\) | Batch-arrival, general service-time queue |
\(M^{X}/G(M/H)/1\) | Unreliable batch-arrival, general service and repair-time after Markovian breakdown |
\(\mathcal {F}_j\) | Maximum allowable vacations to be taken is j |
\(\mathcal {S}\) | Single-vacation policy, i.e., \(j=1\) |
\(\mathcal {M}\) | Multiple-vacation policy, i.e., \(j\rightarrow \infty \) |
\(Q^{\mathcal {P}}(t)\) | Transient number of callers both in service and in orbit under \(\mathcal {P}\) and at t |
\(O^{\mathcal {P}}(t)\) | Transient number of callers in orbit under \(\mathcal {P}\) and at t |
\(I^{\mathcal {P}}(t)\) | Transient number of callers in service under \(\mathcal {P}\) and at t |
\(Q^{\mathcal {F}_j}(z)\) | p.g.f of the system size at arbitrary epoch |
\(V^i_{-}(t)\) | Elapsed \(i^{\text {th}}\) vacation time at t |
\(X^i_{-}(t)\) | Elapsed \(i^{\text {th}}\) service time at t |
\(R^i_{-}(t)\) | Elapsed \(i^{\text {th}}\) repair time at t |
\(C(t)(=\nu , 0,1,2)\) | State of the server at t, i.e., vacation, idle, busy or repair |
\(L^{(k)}\) | k-busy period for the \(M^X/G(M/H)/1\) retrial system without vacation |
\(L^{\mathcal {P}}\) | Busy period of \(M^X/G(M/H)/1\) under \(\mathcal {P}\) and \(L^{\mathcal {P}}=L_0^{\mathcal {P}}+L_w^{\mathcal {P}}+L_r^{\mathcal {P}}\), where |
\(L_0^{\mathcal {P}}\) | Server-waiting time for the next busy epoch under \(L^{\mathcal {P}}\) |
\(L_1^{\mathcal {P}}\) | Generalized working length under \(\mathcal {P}\) (\(L_1^{\mathcal {P}}=L_w^{\mathcal {P}}+L_r^{\mathcal {P}}\)) |
\(L_w^{\mathcal {P}}\) | Value-adding or working length under \(L^{\mathcal {P}}\) |
\(L_r^{\mathcal {P}}\) | Repair length during \(L^{\mathcal {P}}\) |
\(L_0^{(k)}\) | Server-waiting time for the next busy epoch under \(L^{(k)}\) |
\(L_1^{(k)}\) | Generalized working length under k-busy period (\(L_1^{(k)}=L_w^{(k)}+L_r^{(k)}\)) |
\(L_w^{(k)}\) | Value-adding or working length under \(L^{(k)}\) |
\(L_r^{(k)}\) | Repair length during \(L^{(k)}\) |
I(t) | Number of services completed at time t |
\(\omega (t)\) | The elapsed service time of the work at time t |
Proof of Lemma 1
(i) The first part is to compute the p.g.f left by a departing caller. We divide our proofs into two cases.
Case 1 Given that \(Q_n=0\), the server goes into vacation immediately at \(S_n\) because it finds that the orbit is empty. Two events may happen, \(\{Q_n=0,\overline{\nu }=0\}\) and \(\{Q_n=0,\overline{\nu }>0\}\). On the event \(\{Q_n=0,\overline{\nu }=0\}\), we must have \(Q_{n+1}=X-1+M_{n+1}\). On the other hand, for the event \(\{Q_n=0,\overline{\nu }=i\}\) for \(i\ge 1\), we have
Thus, we have
Case 2 For \(Q_n\ge 1\), then we have similar argument. There are two types of customers that competes when \(Q_n\ge 1\). One is the arrival of a new batch of customer, while the other is the random customer from the orbit. On \(\{Q_n=i\}\), where \(i\ge 1\), we have
Thus, we have for \(k\ge i\),
We want to compute the steady state distribution of the system size immediately after departure epoches. Since \(\{Q_n:n\in \mathbf N \}\) constitutes a Markov chain, we can compute its steady state by using \(\pi \mathbf P =\pi \) and \(\pi 1=1\). \(\mathbf P =[p_{ij}]\) is the transition matrix of the embedded Markov Chain \(\{Q_n\}\) and \(\pi =(\pi _0,\pi _1 \pi _2\ldots )\), where \(\pi _k\) is the long run fraction when a departing customer leaves behind k customers in the system. Let \(Q^{\mathcal {F}_j}_{+}(z)\) be the p.g.f of the system size that is left behind by a departing customer, then \(Q^{\mathcal {F}_j}_{+}(z)=\sum _{k=0}^{\infty }\pi _kz^k\). Using (2), (3) and definition of \(Q^{\mathcal {F}_j}_{+}(z)\), we have after some tedious algebraic manipulation,
From the normalization condition, \(Q^{\mathcal {F}_j}_{+}(1)=1\), we can find the constant \(\pi _0\). Using (18) and L’Hospital rule, we obtain
Using (1) and (18), we obtain the p.g.f of the queue length left by a departing customer is given in the main paper.
(ii) (Sufficiency) The sufficiency of the ergodicity result can be shown using Pake’s Lemma which is the statement as follows. If we have \(Q(x)\ge 0\), and for all x, there exist \(\epsilon >0\) such that \(E(Q_{t+1}-Q_t|Q_t=x)\le -\epsilon <0\) for all x except on a finite set C, then \(\{Q_n\}\) is positive recurrent. To show sufficiency of our ergodicity result given any i and \(\kappa (\theta )<1\), we choose \(\epsilon =\frac{1}{2}(1-\kappa (\theta ))>0\). Due to the recursive structure of the embedded sequence \(\{Q_n:n\in \mathbf N \}\), we apply Pake’s lemma as follows. The mean drift \(y_i=E[Q_{n+1}-Q_n|Q_n=i]\) is calculated as follows.
Then, we have \(y_i=-2\epsilon \), implying \(y_i<-\epsilon \) for all states except for a finite number of states. Therefore, \(\kappa (\theta )<1\) is sufficient for the embedded chain to be ergodic. The necessary condition readily follows from Kaplan’s condition [see Senott et al. (1983)], namely \(y_i<\infty \) for all \(i\ge 0\) and there exists \(i_0\in \mathbf Z ^+\) such that \(y_i\ge 0\) for all \(i\ge i_0\). \(\square \)
Proof of Theorem 1
We define the generating functions
For notational parsimony, we omit the superscript \(\mathcal {F}_j\) in the following discussion. Using the technique of supplementary variables, we obtain the following system of equations for \(x\ge 0, y\ge 0, k\ge 0\):
For \(k\ge 1\), we have
These equations are to be solved under the boundary conditions:
The normalizing condition is
Using generating functions, we can express system of equations to be
The boundary conditions can be expressed as
To show \(\omega ^{\mathcal {F}_j}_{\nu ,i}(z,x)\), (30) implies that
Next, (26), we have \(\omega _{\nu ,i}(z,0)=\omega _{\nu ,0,i}(0)\). Using this, we have \(\omega _{\nu ,0,i}(x)=\omega _{i}(0,x)=\omega _{\nu ,0,i}(0)[1-V(x)]e^{-\lambda x}\). Combining it with (25), we have \(\omega _{\nu ,0,i}(0)=\omega _{\nu ,0,i-1}(0)\widetilde{V}(\lambda )\). Recursively, we obtain for \(1\le i\le j\),
In particular, we get
Next, (32) implies
Using both (39) and (36), we get the expression in (6). Substituting (39) into (31), we obtain the following:
From (40), we get \(\omega _1(z,x)=\omega _1(z,0)[1-B(x)]e^{(\lambda -\lambda X(z)+\alpha -\alpha \widetilde{R}(\lambda -\lambda X(z)))x}\). We substitute \(\omega _1(z,x)\),(35), (37), and (38) into (33), we get
After some re-arranging, we obtain
Next using (35) and (42), we have
Integrating (43) w.r.t x, we obtain
Thus, we can obtain \(\omega _2(z)=\alpha \omega _1(z)\frac{1-\widetilde{R}(\lambda -\lambda X(z))}{\lambda -\lambda X(z)}\). From the normalizing condition in (29), we have \(\sum _{i=1}^j\omega _{\nu ,i}(1)+\omega _{0}(1)+\omega _1(1)+\omega _2(1)=1\), and after some tedious algebra, we get \(\omega _{\nu ,0,1}(0)=C_j(\theta )\). \(\square \)
Proof of Corollary 1
We only show (i) as the rest are similar. From Theorem 1, we consider \(\omega ^{\mathcal {F}_j}_{\nu }(1)=\lim _{z\rightarrow 1^-}C_j(\theta )g_j(\lambda )\left( \frac{1-\widetilde{V}(\lambda -\lambda X(z))}{\lambda -\lambda X(z)}\right) \) and applying L’Hospital rule, we obtain \(\omega ^{\mathcal {F}_j}_{\nu }(1)=C_j(\theta )g_j(\lambda ){ EV} =\frac{[1-\kappa (\theta )]g_j(\lambda ){ EV}}{g_j(\lambda ){ EV}+\frac{\theta \widetilde{V}(\lambda )^j}{\lambda (\lambda +\theta )}}< 1\). It is ready to verify that \(\omega ^{\mathcal {F}_j}_{\nu }(1)+\omega ^{\mathcal {F}_j}_0(1)+\omega ^{\mathcal {F}_j}_1(1)+\omega ^{\mathcal {F}_j}_2(1)=1\). \(\square \)
Proof of Corollary 2
Let \(O^{\mathcal {F}_j}(z)\) and \(Q^{\mathcal {F}_j}(z)\) be the p.g.f for the number of callers in orbit and system respectively. In order to prove the results for (i) and (ii), they follow easily from the fact that \(O^{\mathcal {F}_j}(z)=\omega ^{\mathcal {F}_j}_{\nu }(z)+\omega ^{\mathcal {F}_j}_0(z)+\omega ^{\mathcal {F}_j}_1(z)+\omega ^{\mathcal {F}_j}_2(z)\) and \(Q^{\mathcal {F}_j}(z)=\omega ^{\mathcal {F}_j}_{\nu }(z)+\omega ^{\mathcal {F}_j}_0(z)+z(\omega ^{\mathcal {F}_j}_1(z)+\omega ^{\mathcal {F}_j}_2(z))\). \(\square \)
Proof of Theorem 2
We want to show that the p.g.f for the number of customers in the orbit can be written as the sum of three random variables. Let M be the orbit size of the system based on constant retrial policy. First, we observe that Atencia et al. (2008, Theorem 3) has shown that \(N_R=N_0+M\). Finally the p.g.f for \(N^{\mathcal {F}_j}(z)\) allows us to conclude that \(N_{\nu }^{\mathcal {F}_j}=N^{\mathcal {F}_j}+N_0+M\). \(\square \)
Proof of Corollary 7
We have \(EC^{\mathcal {P}}_{\nu }=EL^{\mathcal {P}}+E\zeta _{\nu }^{\mathcal {P}}\) and using results in Theorem 4 and Corollary 6, we have the required result. \(\square \)
Proof of Lemma 2
To show Lemma 2, we consider the following. Denote \(m_n(s)=P\{M=n,\tau _s>H\}\). For any \(k\ge 1\), we have \(\pi _{n,1}^{(k)}(s)=m_{n-k+1}(s)\) and for \(i\ge 2\),
Finally, the proof is completed by showing that \(\sum _{n=0}^{\infty }m_n(s)z^n=\widetilde{H}(s+\lambda -\lambda X(z))\). \(\square \)
For ease of exposition, we define \(F(s,z,y,x)=(s+\lambda +\theta )-\frac{y}{z}(\lambda X(z)+\theta )\widetilde{H}(s+\lambda -\lambda X(z))\).
1.1 Proof of Theorem 3
The proof of Theorem 3 requires two further lemmas, i.e., Lemma 5 and Lemma 6. These results allow us to compute the working and server-waiting length of the k-busy period. Finally, we obtain \(EL^{(k)}\) which agrees with Atencia et al. (2008) who use the technique of supplementary variables.
Lemma 5
The generating function \(\varphi _0^{(k)}(s,z,y)\) satisfies the functional equation
Proof
Using the fact that \(s\varphi ^{(k)}_{0ni}(s)=P\{L^{(k)}>\tau _s,C(\tau _s)=0,Q(\tau _s)=n,I(\tau _s)=i\}=\frac{s}{s+\lambda +\theta }\pi _{ni}^{(k)}(s).\) Thus, we have \(\varphi ^{(k)}_{0ni}(s)= \frac{\pi _{ni}^{(k)}(s)}{s+\lambda +\theta }\). The result follows from (9) since \(\varphi _{0}^{(k)}(s,z,y)=\frac{f^{(k)}(s,z,y)}{s+\lambda +\theta }\). \(\square \)
Lemma 6
The generating function \(\varphi _1^{(k)}(s,z,y,x)\) satisfies the functional equation
In addition, we have
Proof
Observe that \(s\varphi _{1ni}^{(k)}(s,x)=P\{L^{(k)}>\tau _s,C(\tau _s)=1,\omega (\tau _s)\in (x,x+dx),Q(\tau _s)=n,I(\tau _s)=i\}\). Following the arguments in Falin and Templeton (1997) or Artalejo et al. (2002), we obtain
The result follows from the definition of \(\varphi _1^{(k)}(s,z,y,x)\) after tedious algebraic manipulations. Finally, (47) follows from integrating (46) w.r.t x. \(\square \)
Proof of Theorem 3
In order to compute \(L_0^{(k)}\), we need to compute \(\lim _{z\rightarrow 1^-}\varphi (0,z,1)\). Let \(s=0, y=1\) into (45), we obtain
Differentiating the above equation w.r.t z and letting \(z\rightarrow 1\), we obtain the desired result for \(L^{(k)}_0\). The expected generalized working length of the k-busy period is given by \(\lim _{z\rightarrow 1-}\int _0^{\infty }\varphi _1^{(k)}(0,z,1,x)dx\). Note that \(\lim _{z\rightarrow 1^-}\frac{1-\widetilde{H}(\lambda -\lambda X(z))}{\lambda -\lambda X(z)}=EH\). \(\square \)
Proof of Corollary 4
The proof is immediate from Theorem 3 and the fact that \(EH=EB(1+\alpha ER)\). \(\square \)
1.2 Proof of Theorem 4
To prove Theorem 4, we shall begin with a preliminary lemma.
Lemma 7
The generating functions \(\psi _0(s,z,y)\) and \(\psi _1(s,z,y,x)\) satisfy the functional equations
Proof
The results follow from applying Lemmas 5 and 6.
Proof of Theorem 4
From Lemma 48, we let \(s=0, y=1\), we have
Differentiate the above equation w.r.t z and let z approach 1, we obtain \(L_0^{\mathcal {F}_j}\). Combining with (1), \(L_1^{\mathcal {F}_j}\) is obtained by using \(\lim _{z\rightarrow 1^-}\int _0^{\infty }\varphi (0,z,1,x)dx\). Finally, \(EL^{\mathcal {M}}=\varphi (0,1,1)+\lim _{z\rightarrow 1^-}\int _0^{\infty }\varphi (0,z,1,x)dx\).
Proof of Lemma 4
To see this, we apply Theorem 4 and after some re-arranging, we have
Given that \(\widetilde{V}(\lambda )\ge 0\), \(\widetilde{V}(\lambda )^j\) is always increasing function in j. It is easy to see that the sign of \(\frac{\theta }{\lambda (\lambda +\theta )}-\frac{{ EV}}{1-\widetilde{V}(\lambda )}\) determines the if \(EL^{\mathcal {F}_j}\) is increasing or decreasing. \(\square \)
Rights and permissions
About this article
Cite this article
Yeo, W.M., Yuan, XM. & Low, J.M.W. On \(M^{X}/G(M/H)/1\) retrial system with vacation: service helpline performance measurement. Ann Oper Res 248, 553–578 (2017). https://doi.org/10.1007/s10479-016-2207-3
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10479-016-2207-3