Abstract
We consider a discrete-time queueing system where the arrival process is general and each arriving customer brings in a constant amount of work which is processed at a deterministic rate. We carry out a sample-path analysis to derive an exact relation between the set of system size values and the set of waiting time values over a busy period of a given sample path. This sample-path relation is then applied to a discrete-time \(G/D/c\) queue with constant service times of one slot, yielding a sample-path version of the steady-state distributional relation between system size and waiting time as derived earlier in the literature. The sample-path analysis of the discrete-time system is further extended to the continuous-time counterpart, resulting in a similar sample-path relation in continuous time.
Similar content being viewed by others
References
Bruneel, H., Kim, B.G.: Discrete-Time Models for Communication Systems Including ATM. Kluwer, Boston (1993)
Chaudhry, M.L., Kim, N.K.: A complete and simple solution for a discrete-time multi-server queue with bulk arrivals and deterministic service times. Oper. Res. Lett. 31, 101–107 (2003)
Chaudhry, M.L., Yoon, B.K., Chae, K.C.: Waiting-time distribution of a discrete-time multiserver queue with correlated arrivals and deterministic service times: D-MAP/D/\(k\) system. Oper. Res. Lett. 30, 174–180 (2002)
Choi, B.D., Kim, B., Hwang, G.U., Kim, J.-K.: The analysis of a multiserver queue fed by a discrete autoregressive process of order 1. Oper. Res. Lett. 32, 85–93 (2004)
El-Taha, M., Stidham, S.: Sample-Path Analysis of Queueing Systems. Kluwer, Boston (1999)
Gao, P., Wittevrongel, S., Bruneel, H.: Discrete-time multiserver queues with geometric service times. Comput. Oper. Res. 31, 81–98 (2004)
Gao, P., Wittevrongel, S., Bruneel, H.: Delay and partial systems contents for a discrete-time G–D–c queue. 4OR Q. J. Oper. Res. 6, 279–290 (2008)
Gao, P., Wittevrongel, S., Walraevens, J., Moeneclaey, M., Bruneel, H.: Calculation of delay characteristics for multiserver queues with constant service times. Eur. J. Oper. Res. 199, 170–175 (2009)
Heyman, D.P., Stidham Jr, S.: The relation between customer and time averages in queues. Oper. Res. 28, 983–994 (1980)
Humblet, P., Bhargava, A., Hluchyj, M.G.: Ballot theorems applied to the transient analysis of \(n\)D/D/1 queues. IEEE/ACM Trans. Netw. 1, 81–95 (1993)
Hwang, G.U., Choi, B.D., Kim, J.-K.: The waiting time analysis of a queue with a discrete autoregressive model of order 1. J. Appl. Probab. 39, 619–629 (2002)
Hwang, G.U., Sohraby, K.: On the exact analysis of a discrete-time queueing system with autoregressive inputs. Queueing Syst. 43, 29–41 (2003)
Knuth, D.E.: The Art of Computer Programming, vol. 2, 3rd edn. Addison-Wesley, Reading (1997)
Little, J.D.C.: Little’s law as viewed on its 50th anniversary. Oper. Res. 59, 536–549 (2011)
Robertazzi, T.G.: Computer Networks and Systems: Queueing Theory and Performance Evaluation. Springer, New York (1994)
Sakasegawa, H., Wolff, R.W.: The equality of the virtual delay and attained waiting time distributions. Adv. Appl. Probab. 22, 257–259 (1990)
Sengupta, B.: An invariance relationship for the G/G/1 queue. Adv. Appl. Probab. 21, 956–957 (1989)
Sigman, K.: A note on a sample-path rate conservation law and its relationship with \(H=\lambda G\). Adv. Appl. Probab. 23, 662–665 (1991)
Stidham, S.: A last word on \(L = \lambda W\). Oper. Res. 22, 417–421 (1974)
Stidham, S.: Analysis, design, and control of queueing systems. Oper. Res. 50, 197–216 (2002)
Takagi, H.: Queueing Analysis, A Foundation of Performance Evaluation, Vol. 3: Discrete-Time Systems. North-Holland, Amsterdam (1993)
Whitt, W.: A review of \(L = \lambda W\) and extensions. Queueing Syst. 9, 235–268 (1991)
Wolff, R.W.: Little’s law and related results. In: Cochran, J.J. (ed.) Wiley Encyclopedia of Operations Research and Management Sciences, vol. 4, pp. 2828–2841. Wiley, Chichester (2011)
Woodward, M.E.: Communication and Computer Networks: Modelling with Discrete-Time Queues. Pentech Press, London (1993)
Xiong, Y., Bruneel, H., Steyaert, B.: Deriving delay characteristics from queue length statistics in discrete-time queues with multiple servers. Perform. Eval. 24, 189–204 (1996)
Yamazaki, G., Miyazawa, M.: The equality of the workload and total attained waiting time in average. J. Appl. Probab. 28, 238–244 (1991)
Yao, Y.-C.: A duality relation between workload and attained waiting time in FCFS G/G/s queues. J. Appl. Probab. 50, 300–307 (2013)
Acknowledgments
The authors are grateful to the two reviewers for helpful comments which led to an improved presentation. The first author acknowledges support from the National Science Council of Taiwan. The second author acknowledges the support from Academia Sinica, Taiwan for his visits to the Institute of Statistical Science.
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
1.1 Lemmas and proofs
In this appendix, we state and prove a few lemmas that are needed for the proof of Theorem 1 in Sect. 2.
Lemma 1
For the discrete-time \((c,k)\)-system, let \(Q=\{(t_1,d_1),\dots ,(t_n,d_n)\}\) be a \(B(n)\) set, and let \(P=\{(t_1,d_1),\dots ,(t_n,d_n),(t_{n+1},d_{n+1})\}\) with \(t_{n+1}=d_n\) and \(d_{n+1}=t_1+\lceil k(n+1)/c\rceil -1\) \((\)which is a \(B(n+1)\) set \()\). If \(S^{[c]}_Q=W^{[k]}_Q\), then \(S^{[c]}_P=W^{[k]}_P\).
Proof
Consider the setting where the \((c,k)\)-system is empty right before \(t_1\), and \(n+1\) customers \(\mathcal C _1,\dots ,\mathcal C _{n+1}\) enter the system at \(t_1,\dots ,t_{n+1}\), respectively. Since \(P\) is a \(B(n+1)\) set, the \(\mathcal C _i\) depart at \(d_i=t_1+\lceil ki/c\rceil -1\), \(i=1,\dots ,n+1\), and exactly \(c\) units of work is performed at each of the slots \(t_1,\dots ,d_{n+1}-1\). Let \(\ell =N_Q(d_n)=|\{i\le n:d_i=d_n\}|\), the number of those in \(\mathcal C _1,\dots ,\mathcal C _n\) who are in system at \(d_n\) (and also depart at the end of slot \(d_n\)), and \(v=\) the number of units of work remaining for these \(\ell \) customers at (the beginning of) slot \(d_n\). We have (cf. (2.3)),
Also \(N_P(t)=N_Q(t)\) for \(t_1\le t<d_n\), since \(t_{n+1}=d_n\). We consider the three cases (\(c\ge 1,k=1\)), (\(c=1,k\ge 1\)) and (\(c\ge 2, k\ge 2\)) separately.
Case (i): \(c\ge 1,k=1\). By (6.1), \(\ell =\lceil v/k\rceil =v\). Write \(n=qc+r\) with \(q=\lceil n/c\rceil -1\ge 0\) and \(0<r\le c\). With \(k=1\), the total number of work units for customers \(\mathcal C _1,\dots ,\mathcal C _n\) is \(n\). Since exactly \(c\) units of work is performed at \(t\in \{t_1,\dots ,d_n-1\}\) and since \(\mathcal C _n\) departs at \(d_n\), we must have \(r=v=\ell \). Two subcases are considered below.
Subcase (i.1): \(0<r<c\). In this case, the total number of work units (including the work of \(\mathcal C _{n+1}\)) at \(d_n\) equals \(r+1\le c\), so that \(d_{n+1}=d_n\). It follows that
Since \(\{r\}^{[c]}=\{1_{(r)}\}\) and \(\{r+1\}^{[c]}=\{1_{(r+1)}\}\) and \(S^{[c]}_Q=W_Q^{[k]}=W_Q\),
Subcase (i.2): \(r=c\). The total number of work units (including the work of \(\mathcal C _{n+1}\)) at \(d_n\) equals \(c+1\), so that \(d_{n+1}=d_n+1\). It follows that
Since \(\{1,c+1\}^{[c]}=\{1_{(c)},2\}\) and \(\{c\}^{[c]}=\{1_{(c)}\}\) and \(S^{[c]}_Q=W_Q\),
This completes the proof for case (i).
Case (ii): \(c=1,k\ge 1\). Since \(c=1\), all of \(\mathcal C _1,\dots ,\mathcal C _{n-1}\) must have departed before \(d_n\), and \(\mathcal C _n\) has just one unit of work remaining at \(d_n\). So \(\ell =v=1\). When \(\mathcal C _{n+1}\) enters the system at \(t_{n+1}=d_n\), it takes \(k\) additional slots to complete the work of \(\mathcal C _{n+1}\). So \(d_{n+1}=d_n+k\). It follows that
Since \(\{k+1\}^{[k]}=\{1_{(k-1)},2\}\) and \(S_Q=S_Q^{[c]}=W^{[k]}_Q\),
completing the proof for case (ii).
Case (iii): \(c\ge 2,k\ge 2\). We further consider the following three subcases.
Subcase (iii.1): \(d_{n+1}=d_n\). Necessarily, \(c\ge v+k\) (the amount of work at \(d_n\), including the work of \(\mathcal C _{n+1}\)), which together with (6.1) implies \(c>\ell k\ge 2\ell \). Also we have \(W_P=W_Q\uplus \{1\}\), \(S_Q=\{N_Q(t):t_1\le t<d_n\}\uplus \{\ell \}\), and \(S_P=\{N_Q(t):t_1\le t<d_n\}\uplus \{\ell +1\}\). Since \(\ell +1\le 2\ell <c\), we have \(\{\ell \}^{[c]}=\{1_{(\ell )}\}\), \(\{\ell +1\}^{[c]}=\{1_{(\ell +1)}\}\), so that
Subcase (iii.2): \(d_{n+1}=d_n+1\). Necessarily, \(v+k\le 2c\), which together with (6.1) implies \(2c\ge v+k>\ell k\ge 2\ell \), i.e. \(c>\ell \). Then \(W_P=W_Q\uplus \{2\}\) (implying \(W^{[k]}_P=W^{[k]}_Q\uplus \{1_{(2)}\}\)),
from which and \(c>\ell \) it follows that
Subcase (iii.3): \(d_{n+1}\ge d_n+2\). Necessarily \(\ell =1\). Since \(\mathcal C _{n+1}\) is the only customer in system after \(d_n\) and since exactly \(c\) units of work is performed at each of the slots \(d_n+1,\dots ,d_{n+1}-1\), we have \(k\ge c(d_{n+1}-d_n-1)+1\ge d_{n+1}-d_n+1\). So \(W_P=W_Q\uplus \{d_{n+1}-d_n+1\}\) (implying \(W^{[k]}_P=W^{[k]}_Q\uplus \{1_{(d_{n+1}-d_n+1)}\}\)),
It follows that
This completes the proof for case (iii). The proof of Lemma 1 is complete. \(\square \)
Lemma 2
For the discrete-time \((c,k)\)-system, let \(Q=\{(t_1,d_1),\dots ,(t_n,d_n)\}\) be a \(B(n)\) set with \(t_n<d_n\). Let \(j\) satisfy \(t_n\le j<j+1\le d_n\), and let \(P=\{(t_1,d_1),\dots ,(t_n,d_n),(j,d_{n+1})\}\) and \(P^{\prime }=\{(t_1,d_1),\dots ,(t_n,d_n),(j+1,d_{n+1})\}\) \((\)with \(d_{n+1}=t_1+\lceil (n+1)k/c\rceil -1)\), which are both \(B(n+1)\) sets. Then
Proof
Consider the setting where the \((c,k)\)-system is empty right before \(t_1\), and \(n+1\) customers \(\mathcal C _1,\dots ,\mathcal C _{n+1}\) enter the system at \(t_1,\dots ,t_n,t_{n+1}\), respectively, where \(t_{n+1}=j\) or \(j+1\). Since \(P\) and \(P^{\prime }\) are both \(B(n+1)\) sets, the \(\mathcal C _i\) depart at \(d_i=t_i+\lceil ki/c\rceil -1\), \(i=1,\dots ,n+1\), and exactly \(c\) units of work is done at each of the slots \(t_1,\dots ,d_{n+1}-1\). Let \(\ell =N_Q(j)\), the number of those in \(\mathcal C _1,\dots ,\mathcal C _n\) who are in system at \(j\), and \(v=\) the number of units of work remaining at \(j\) for the \(\ell \) customers \(\mathcal C _{n-\ell +1},\dots ,\mathcal C _n\). Since none of \(\mathcal C _{n-\ell +2},\dots ,\mathcal C _n\) begins service before time \(j\), we have \((\ell -1)k+1\le v\le \ell k\), implying \(\ell =\lceil v/k\rceil \) (cf. (2.3)). On the other hand, with \(t_{n+1}=j\), \(v+k\) units of work (including \(\mathcal C _{n+1}\)’s work) needs to be completed at \(d_{n+1}\), so \((d_{n+1}-j)c+1\le v+k\le (d_{n+1}-j+1)c\), implying \(d_{n+1}-j+1=\lceil \frac{v+k}{c}\rceil \). We have shown
Since \(\mathcal C _{n+1}\) departs at \(d_{n+1}\) for both cases \(t_{n+1}=j\) and \(t_{n+1}=j+1\), the waiting time of \(\mathcal C _{n+1}\) is \(d_{n+1}-j+1\) for the former case and \(d_{n+1}-j\) for the latter. As the waiting times of \(\mathcal C _1,\dots ,\mathcal C _n\) do not depend on \(t_{n+1}\), we have
Also, \(\mathcal C _{n+1}\) begins service at the same time for the two cases \(t_{n+1}=j\) and \(t_{n+1}=j+1\). In particular, for \(t_{n+1}=j\), \(\mathcal C _{n+1}\) cannot begin service at \(j (<d_n)\). It follows that
So,
By Lemma 3 below, \(\lceil \lceil (v+k)/k\rceil /c\rceil =\lceil \lceil (v+k)/c\rceil /k\rceil \), which is denoted by \(\rho \). We consider \(\rho =1\) and \(\rho >1\) separately.
Case (i): \(\rho =1\). We have \(\lceil (v+k)/k\rceil \le c\) and \(\lceil (v+k)/c\rceil \le k\), so
It follows that \(S^{[c]}_P=W^{[k]}_P\) if and only if \(S^{[c]}_{P^{\prime }}=W^{[k]}_{P^{\prime }}\).
Case (ii): \(\rho >1\). It is readily seen that \(\{\lceil \frac{v+k}{k}\rceil \}^{[c]}\) has one more copy of \(\rho \) and one fewer copy of \(\rho -1\) than \(\{\lceil \frac{v+k}{k}\rceil -1\}^{[c]}\), and \(\{\lceil \frac{v+k}{c}\rceil \}^{[k]}\) has one more copy of \(\rho \) and one fewer copy of \(\rho -1\) than \(\{\lceil \frac{v+k}{c}\rceil -1\}^{[k]}\). It follows from (6.5) and (6.6) that
which implies that \(S^{[c]}_P=W^{[k]}_P\) if and only if \(S^{[c]}_{P^{\prime }}=W^{[k]}_{P^{\prime }}\). This completes the proof. \(\square \)
Lemma 3
For positive integers \(i,j\) and \(\ell \), we have
Proof
Let \(\rho =\lceil \ell /i\rceil \). Then \(\ell \le i\rho \), and \(\lceil \ell /j\rceil \le \lceil i\rho /j\rceil \le i\lceil \rho /j\rceil \), implying \(\lceil \lceil \ell /j\rceil /i\rceil \le \lceil \rho /j\rceil =\lceil \lceil \ell /i\rceil /j\rceil \). By symmetry, \(\lceil \lceil \ell /i\rceil /j\rceil \le \lceil \lceil \ell /j\rceil /i\rceil \). This completes the proof. \(\square \)
Rights and permissions
About this article
Cite this article
Yao, YC., Miao, D.WC. Sample-path analysis of general arrival queueing systems with constant amount of work for all customers. Queueing Syst 76, 283–308 (2014). https://doi.org/10.1007/s11134-013-9361-y
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11134-013-9361-y