Abstract
In this paper, we consider a queue with impatient customers, under general assumptions. We introduce a convenient representation of the system by a stochastic recursive sequence keeping track of the remaining service and patience times of the customers. This description allows us (i) to provide a comprehensive stability condition in the general case, (ii) to give a rigorous proof of the optimality of the Earliest Deadline First (EDF) service discipline in several cases, and (iii) to show that the abandonment probability of the system follows inversely the stochastic ordering of the generic patience time distribution.
Similar content being viewed by others
Notes
Notice that the result presented in [1] only concerns GI/GI/1 queues. However, it should be clear that this also holds true whenever the inter-arrival times are not independent of one another, but only independent of the service times. Indeed, even in that case any permutation of the service times clearly do not affect the global statistics of the system, provided that the arrival times and patience times are left unchanged, and that the service discipline is independent of the service times.
References
Baccelli, F., Brémaud, P.: Elements of Queueing Theory, 2nd edn. Springer, Berlin (2002)
Baccelli, F., Hébuterne, G.: On queues with impatient customers. In: Performance’81, pp. 159–179 (1981)
Baccelli, F., Boyer, P., Hébuterne, G.: Single-server queues with impatient customers. Adv. Appl. Probab. 16(4), 887–905 (1984)
Barrer, D.Y.: Queuing with impatient customers and ordered service. Oper. Res. 5, 650–656 (1957)
Borovkov, A.A., Foss, S.: Stochastic recursive sequences and their generalizations. Sib. Adv. Math. 2(1), 16–81 (1992)
Borovkov, A.A., Foss, S.: Two ergodicity criteria for stochastically recursive sequences. Acta Appl. Math. 34, 125–134 (1994)
Decreusefond, L., Moyal, P.: The fluid limit of a heavily loaded EDF queue with impatient customers. Markov Process. Relat. Fields 14(1), 131–158 (2008)
Decreusefond, L., Moyal, P.: Stochastic Modeling and Analysis of Telecom Networks. Wiley-ISTE (2012)
Decreusefond, L., Moyal, P.: A functional central limit theorem for the M/GI/\(\infty \) queue. Ann. Appl. Probab. 18(6), 2156–2178 (2008)
Dertouzos, M.: Control robotics: the procedural control of physical processus. Proc. IFIP Congress (1974)
Doytchinov, B., Lehoczky, J.P., Shreve, S.: Real-time queues in heavy-traffic with earliest deadline first queue discipline. Ann. Appl. Probab. 11(2), 332–378 (2001)
Flipo, D.: Comparaison des disciplines de service des files d’attente G/G/1 (in french). Annales de l’institut Henri Poincaré, Section B 17(2), 190–197 (1981)
Gromoll, C., Puha, A., Williams, R.: The fluid limit of a heavily loaded processor sharing queue. Ann. Appl. Probab. 12, 797–859 (2002)
Gromoll, C., Robert, Ph., Zwart, B.: Fluid limits for processor sharing queues with impatience. Math. Oper. Res. 33, 375–402 (2008). (2006), no. 34(1)
Jackson, J.R.: Scheduling a production line to minimize maximum tardiness. Res. Rep. 43, Management Sci. Rep. Univ. of Calif., Los Angeles (1955)
Kruk, L., Lehoczky, J., Ramanan, K., Shreve, S.: Heavy traffic analysis for EDF queues with reneging. Ann. Appl. Probab. 21(2), 484–545 (2011)
Liu, C.L., Layland, J.W.: Scheduling algorithms for multiprogramming in a hard real-time environment. J. Assoc. Comput. Mach. 20(1), 46–61 (1973)
Loynes, R.M.: The stability of queues with non-independent interarrivals and service times. Proce. Cambridge Philos. Soc. 58, 497–520 (1962)
Moyal, P.: Stability of a processor sharing queue with varying throughput. Journ. Appl. Probab. 45(4), 953–962 (2008)
Moyal, P.: Convex comparison of service disciplines in Real-time queues. Oper. Res. Lett. 36(4), 496–499 (2008)
Moyal, P.: The queue with impatience: construction of the stationary workload under FIFO. J. Appl. Probab. 47(4), 498–512 (2010)
Panwar, S.S., Towsley, D.: Optimality of the stochastic earliest dead- line policy for the G/M/c queue serving customers with deadlines. Second ORSA Telecommunications Conference, Boca Raton (March 1992)
Panwar, S.S., Towsley, D., Wolf, J.K.: Optimal scheduling policies for a class of queues with customer deadlines to the beginning of service. J. ACM. 35(4), 832–844 (1988)
Pinedo, M.: Stochastic scheduling with release dates and due dates. Oper. Res. 31, 559–572 (1983)
Sorenson, P.G., Hamacher, V.C.: A real-time system design methodology. INFOR 13(1), 1–18 (1975)
Stoyenko, A.D., Georgiadis, L.: On optimal lateness and tardiness scheduling in real-time systems. Computing 47, 215–234 (1992)
Thorisson, H.: Coupling, Stationarity and Regeneration. Springer, Berlin (2000)
Author information
Authors and Affiliations
Corresponding author
Appendix: Technical results
Appendix: Technical results
We first have the following result.
Lemma 2
Let \(u \in \mathcal{S }^2\). Assume that \(\underline{u}\) is such that
Denote \(\alpha \), the permutation of \([[0,N(u)-1 ]]\) such that
Let \(\hat{\underline{u}}\) be the transformation of \(\underline{u}\) w.r.t. \(u\), as defined above. Let \(x>0\). Denote \(p:=p(u,\,x)=\text{ Card} \mathcal{A }(u,\,x)-1\) (respectively \(q:=q(\hat{\underline{u}},\,x)=\text{ Card} \mathcal{A }(\hat{\underline{u}},\,x)-1\)) and
Then, there exists an injection
Proof
We start by showing that for all \(\ell \in [[1,p\wedge q ]]\), there exists a bijection
For this, we proceed by induction. First, for \(\ell =0\) the property clearly holds if \(\alpha (0)=0\). If not, as \(\underline{u}^2\) is ranked in decreasing order,
so we can set \(F_\ell (i_0)=k_0\).
Assume now that the result holds for \(\ell -1\), where \(\ell \in [[1,p\wedge q ]]\). There are two possible cases:
-
(i)
if \(k_\ell \le \alpha (i_\ell )\), then
$$\begin{aligned} \underline{u}^2\left(N(u)-k_\ell \right) \le \underline{u}^2\left(N(u)-\alpha (i_\ell )\right)=u^2\left(N(u)-i_\ell \right), \end{aligned}$$and it suffices to set \(F_{\ell }(i_m)=F_{\ell -1}(i_m)\) for all \(m \le \ell -1\) and \(F_{\ell }(i_\ell )=k_\ell \).
-
(ii)
if \(k_\ell > \alpha (i_\ell )\), we have
$$\begin{aligned} x \wedge \underline{u}^2\left(N(u)-\alpha (i_\ell )\right)&= x \wedge u^2\left(N(u)-i_\ell \right)\\&> \sum _{j=1}^{\ell -1} u^1\left(N(u)-i_j\right)\\&= \sum _{j=1}^{\ell -1} \hat{\underline{u}}^1\left(N(u)-k_j\right)\\&\ge \sum _{j=1}^{\ell -2} \hat{\underline{u}}^1\left(N(u)-k_j\right). \end{aligned}$$Therefore, \(\alpha (i_\ell ) \in \mathcal{A }^{k_{\ell -1}}\left(\underline{u},\,x\right),\) so \(\alpha (i_\ell )=F_{\ell -1}(i_j)\) for some \(j\in [[0,\ell -1 ]]\). We have
$$\begin{aligned} x \wedge \underline{u}^2\left(N(u)-\alpha (i_j)\right)&= x \wedge u^2\left(N(u)-i_j\right) \nonumber \\&\ge x \wedge \underline{u}^2\left(N(u)-F_{\ell -1}(i_j)\right) \nonumber \\&= x \wedge \underline{u}^2\left(N(u)-\alpha (i_\ell )\right)\nonumber \\&= x \wedge u^2\left(N(u)-i_\ell \right)\nonumber \\&> \sum _{j=1}^{\ell -1} u^1\left(N(u)-i_j\right)\nonumber \\&= \sum _{j=1}^{\ell -1} \hat{\underline{u}}^1\left(N(u)-k_j\right). \end{aligned}$$(45)Furthermore, as \(i_\ell \not \in \mathcal{A }^{i_{\ell -1}}\left(u,\,x\right),\) it is clear by the very construction of \(F_{\ell -1}\) that \(\alpha (i_j) \not \in \mathcal{A }^{k_{\ell -1}}\left(\underline{u},\,x\right)\), so from (45), \(\alpha (i_j)>k_{\ell -1}\). But by definition of \(\underline{u}\) and \(k_\ell \),
$$\begin{aligned} k_\ell =\min \biggl \{q \in [[k_{\ell -1}+1,N(u)-1 ]]; x \wedge \underline{u}^2\left(N(u)-q\right) > \sum _{j=1}^{\ell -1} \hat{\underline{u}}^1\left(N(u)-k_j\right)\biggl \}. \end{aligned}$$Therefore, we deduce from (45) that
$$\begin{aligned} \alpha (i_j)\ge k_\ell . \end{aligned}$$(46)There are three subcases:
-
(a)
If \(k_\ell =\alpha (i_j)\), we can set
$$\begin{aligned} \left\{ \begin{array}{ll} F_{\ell }(i_\ell )&=\alpha (i_\ell );\\ F_{\ell }(i_j)&=k_\ell ;\\ F_{\ell }(i_m)&=F_{\ell -1}\left(i_m\right),\, \text{ for} \text{ all} m \in [[0,\ell -1 ]]\setminus \{j\}. \end{array} \right. \end{aligned}$$ -
(b)
If \(k_\ell =\alpha (i_p)\) for some \(p \in \mathcal{A }^{i_{\ell -1}}\left(u,\,x\right),\,p\ne j\), in view of (46) we have
$$\begin{aligned} \underline{u}^2\left(N(u)-F_{\ell -1}(i_p)\right)&\le u^2\left(N(u)-i_p\right)\\&= \underline{u}^2\left(N(u)-k_\ell \right)\\&\le \underline{u}^2\left(N(u)-\alpha (i_j)\right)\\&= u^2\left(N(u)-i_j\right). \end{aligned}$$Therefore, we can set
$$\begin{aligned} \left\{ \begin{array}{ll} F_{\ell }(i_\ell )&=\alpha (i_\ell );\\ F_{\ell }(i_p)&=k_\ell ;\\ F_{\ell }(i_j)&=F_{\ell -1}\left(i_p\right);\\ F_{\ell }(i_m)&=F_{\ell -1}\left(i_m\right),\, \text{ for} \text{ all} m \in [[0,\ell -1 ]]\setminus \{j,p\}. \end{array} \right. \end{aligned}$$ -
(c)
Finally, in the other cases \(\alpha (i_j) \not \in \mathcal{A }^{k_{\ell }}(\underline{u},\,x)\) and \(\underline{u}^2\left(N(u)-k_\ell \right)\le u^2\left(N(u)-i_j\right)\), so we can set again
$$\begin{aligned} \left\{ \begin{array}{ll} F_{\ell }(i_\ell )&=\alpha (i_\ell );\\ F_{\ell }(i_j)&=k_\ell ;\\ F_{\ell }(i_m)&=F_{\ell -1}\left(i_m\right),\, \text{ for} \text{ all} m \in [[0,\ell -1 ]]\setminus \{j\}. \end{array} \right. \end{aligned}$$
-
(a)
It follows from the existence of \(F_p\), that \(p\le q\). Indeed, if we assume that some index \(i_{q+1}\) belongs to \(\mathcal{A }\left(u,\,x\right)\), then
Therefore, \(\alpha \left(i_{q+1}\right) \in \mathcal{A }(\underline{u},\,x)\), say \(\alpha \left(i_{q+1}\right)=F_{\ell }(i_{j}),\) where \(i_j \in \mathcal{A }^{i_\ell }(u,\,x)\). So by the very construction of the mapping \(F_\ell ,\,\alpha (i_j) \not \in \mathcal{A }^{k_\ell }(\underline{u},\,x)\), whereas
Therefore, there exists an index
a contradiction. This shows that \(q \ge p.\)
The proof of the Lemma is thus completed, by setting \(F \equiv F_{p}\), which is injective by construction. \(\square \)
Lemma 3
Let \(u\) and \(v \in \mathcal{S }^2\) such that \(u^2\) and \(v^2\) are ordered in decreasing order, and such that
Let \(x>0\). Denote \(q:=q(u,\,x)=\text{ Card} \mathcal{A }(u,\,x)-1\) (respectively \(r:=r(\hat{v},\,x)=\text{ Card} \mathcal{A }(\hat{v},\,x)-1\)) and
where \(\hat{v}\) is the modified version of \(v\) with respect to \(u\) as defined above. Then, \(q \le r\) and for any \(\ell \in [[0,q ]]\),
Proof
We first show by induction that (48) holds for any \(\ell \in [[0,q\wedge r ]]\). It is clear in view of (47) that \(N(u) \le N(v)\), so (48) holds for \(\ell =0\). Suppose that it holds true for \(\ell -1\) (with \(\ell \in [[1,q\wedge r ]]\)). We have
But in view of the recurrence assumption,
Therefore, by the very definition of \(l_\ell \), we deduce from (49) that
which completes the recurrence.
It remains to show that \(q \le r\). Assume that this does not hold, i.e., some index \(k_{r+1} \in \mathcal{A }(u,\,x)\). This implies that
As the term on the right-hand side is larger that the latter sum up to any \(\ell \le r\), this implies for some \(\ell \le r,\, N(u)-k_{r+1}=N(v)-l_\ell . \) But in view of the previous result, this implies that
a contradiction. Thus \(q \le r\). \(\square \)
Lemma 4
Let \(p \le q\) and \(u \in \mathbf{R }^p,\,v\in \mathbf{R }^q\). Assume that there exists an injection \(\gamma :\,[[1,p ]]\rightarrow [[1,q ]]\) such that
Thus,
Proof
Denote \(\delta \) and \(\varepsilon \) the respective permutations of \([[1,p ]]\) and \([[1,q ]]\) such that
We proceed by induction. First,
Assume that the property holds until \(j\). First, if
then \(\gamma \circ \delta (j+1)=\varepsilon (\ell )\) for some \(\ell >j\), so in particular
If now \(\gamma \circ \delta (j+1)\in \varepsilon \left([[1,j ]]\right)\), as \(\gamma \circ \delta \) is injective, there exist \(k \le j\) and \(m > j\) such that
So we have that
which completes the proof. \(\square \)
This finally leads to the following result.
Theorem 5
Let \(u\) and \(v\) be two elements of \(\mathcal{S }^2\) such that \(v^2\) is ordered in decreasing order. Denote \(\hat{v}\) and \(\hat{\underline{u}}\), the modified versions of \(v\) and \(\underline{u}\) with respect to \(u\). Assume that
Let \(x>0\). Then,
and if \(H_2\left(u,\,x\right)\ne \mathbf 0 \), we have
where the mapping \(H_2\) is defined by (16).
Proof
The result holds true whenever \(H_2\left(u,\,x\right)=\mathbf 0 \), which is the case in particular if \(u^2=\mathbf 0 _2\) since \(\mathcal{A }(u,\,x)=\{0\}\) by construction.
If not, we consider the sequences \(u,\underline{u}\) and \(v\) and apply Lemmas 2, 3 and 4. Here again, we denote \(\alpha \) the permutation such that
and \(F\), the injection defined in Lemma 2. We denote as well
with \(p \le q \le r.\)
By the very definition of \(H_2\), any positive second component of \(H_2(u,\,x)\) reads \(u^2\left(N(u)-i\right)-x\), where \(i\in [[i_p+1,N(u)-1 ]]\) and
We then necessarily have that
because the contrary would imply that
so some index \(i_{p+1} \in [[i_p+1,i ]]\) would belong to \(\mathcal{A }(u,\,x)\), a contradiction. Therefore, (51) implies that \(r \le p\), so we have that
and in particular, \(F\) defines a bijection from \(\mathcal{A }(u,\,x)\) to \(\mathcal{A }(\hat{\underline{u}},\,x)\). Furthermore, this implies that
hence (50) is verified. Now, there are two cases:
-
(i)
if \(\alpha (i) \not \in \mathcal{A }\left(\hat{\underline{u}},\,x\right)\), we have \(\alpha (i)> k_p\) and
$$\begin{aligned} \underline{u}^2\left(N(u)-\alpha (i)\right)=u^2\left(N(u)-i\right)>x> \sum _{j=1}^{p-1} \hat{\underline{u}}^1\left(N(u)-k_j\right), \end{aligned}$$so some positive component of \(\left(H_2\left(\hat{\underline{u}},\,x\right)\right)^2\) reads
$$\begin{aligned} \underline{u}^2\left(N(u)-\alpha (i)\right)-x=u^2\left(N(u)-i\right)-x. \end{aligned}$$(53) -
(ii)
if \(\alpha (i) \in \mathcal{A }\left(\hat{\underline{u}},\,x\right),\) then \(\alpha \circ F^{-1} \circ \alpha (i) > k_p\) and we have that
$$\begin{aligned} \underline{u}^2\left(N(u)-\alpha \circ F^{-1} \circ \alpha (i)\right)&= u^2\left(N(u)- F^{-1} \circ \alpha (i)\right)\\&\ge \underline{u}^2\left(N(u)- \alpha (i)\right)\\&= u^2\left(N(u)-i\right)\\&> x >\sum _{j=1}^{p-1} \hat{\underline{u}}^1\left(N(u)-k_j\right), \end{aligned}$$thus some positive component of \(\left(H_2\left(\hat{\underline{u}},\,x\right)\right)^2\) is given by
$$\begin{aligned} \underline{u}^2\left(N(u)-\alpha \circ F^{-1} \circ \alpha (i)\right)-x \ge u^2\left(N(u)-i\right)-x. \end{aligned}$$(54)
Now, any positive second component of \(H_2\left(\hat{\underline{u}},\,x\right)\) reads \(\underline{u}^2\left(N(u)-j\right)-x\), where \(j \in [[k_p+1,N(u) ]]\). Assume that \(N(v)-N(u)+j \in \mathcal{A }(\hat{v},\,x)\) and denote the index \(\ell \in [[0,p ]]\) such that \(N(v)-N(u)+j=l_\ell \). From Lemma 3, \(N(u)-k_\ell \le N(v)-l_\ell ,\) thus \(N(u)-k_p \le N(u)-j\), a contradiction. Therefore, \(N(v)-N(u)+j \not \in \mathcal{A }(\hat{v},\,x)\). But as
the index \(N(v)-N(u)+j\) is strictly larger than \(l_p\). This means that some positive second component of \(v\) reads
Therefore, applying (53), (54) and (55) to any index \(i \in [[i_p+1,N(u)-1 ]]\), and to \(j=\alpha (i)\) in the case (i) and to \(j=\alpha \circ F^{-1} \circ \alpha (i)\) in the case (ii), we obtain that the number \(a\) of positive components of \(\left(H_3(u,\,x)\right)^2\) is less than the number \(b\) of components of \(\left(H_3(\hat{v},\,x)\right)^2\), and that there exists an injection \(\gamma :[[1,a ]]\rightarrow [[1,b ]]\) such that for all \(i\),
We conclude with Lemma 4, to obtain that
which completes the proof. \(\square \)
Rights and permissions
About this article
Cite this article
Moyal, P. On queues with impatience: stability, and the optimality of Earliest Deadline First. Queueing Syst 75, 211–242 (2013). https://doi.org/10.1007/s11134-013-9342-1
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11134-013-9342-1