Matrix geometric analysis of the shortest queue problem with
threshold jockeying
Citation for published version (APA):
Adan, I. J. B. F., Wessels, J., & Zijm, W. H. M. (1991). Matrix geometric analysis of the shortest queue problem with threshold jockeying. (Memorandum COSOR; Vol. 9124). Technische Universiteit Eindhoven.
Document status and date: Published: 01/01/1991
Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)
Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please follow below link for the End User Agreement:
www.tue.nl/taverne
Take down policy
If you believe that this document breaches copyright please contact us at:
openaccess@tue.nl
providing details and we will investigate your claim.
Memorandum COSOR 91-24 Matrix-geometric analysis of the shortest
queue problem with threshold jockeying I.J.B.F. Adan
J. Wessels W.H.M. Zijm
Eindhoven University of Technology
Department of Mathematics and Computing Science P.O. Box 513
5600 MB Eindhoven The Netherlands
Eindhoven, October 1991 The Netherlands
MATRIX-GEOMETRIC ANALYSIS OF THE SHORTEST QUEUE PROBLEM WITH THRESHOLD JOCKEYING
IJ.BF. Adanl), J. Wessels1).2), W.HM. Zijm3)
Abstract: In this paper we study a system consisting ofcparallel servers with possibly dif-ferent service rates. Jobs arrive according toa Poisson stream and generate an exponen-tially distributed workload. An arriving job joins the shortest queue, where in case of multi-ple shortest queues, one of these queues is selected accordingtosome arbitrarY probability distribution.Ifthe maximal difference between the lengths of thec queues exceeds some threshold valueT, then one job switches from the longesttothe shortest queue, where in case of multiple longest queues, the queue loosing a job is selected accordingtosome
arbi-traryprobability distribution. It is shown that the matrix-geometric approach is very well suitedtofind the equilibrium probabilities of the queue lengths. The interesting point is that for one partitioning of the state space an explicit ergodicity condition canbederived from Neuts' mean drift condition, whereas for another partitioning the associated
R
-matrix can bedetermined explicitly. Moreover, both partitionings used are different from the one sug-gested by the conventional way of applying the matrix-geometric approach. Therefore, the paper can beseen as a plea for giving more attention to the question of the selection of a partitioning in the matrix-geometric approach.Key Words: jockeying, matrix-geometric solution, queues in parallel, shortest queue.
1. Introduction
The matrix-geometric approach, as introduced by Neuts in his book [8], has proved to be a
powerful tool for the analysis of markov processes with large and complicated state spaces,
par-ticularly the ones that appear when modeling queueing or maintenance systems. One stage in
the approach is a partitioning of the state space. Usually this stage does not get much attention.
Users tend to use so-called natural partitionings which are suggested by the way of modeling or,
if such a natural partioning is not available, they select a partitioning that fits most elegantly
with the boundary behaviour of the process.
Ina previous paper
[1],the authors demonstrate
already that in the case of the shortest queue problem with threshold jockeying for 2 parallel
queues it is more effective to apply a partitioning based on the behaviour of
theprocess in the
1) Eindhoven University of Technology, Department of Mathematics and Computer Science, p.o. Box 513,5600 MB - Eindhoven, The Netherlands.
2) International Institute for Applied Systems Analysis, A-2361 Laxenburg, Austria.
3) University of Twente, Department of Mechanical Engineering, p.o. Box 217, 7500 AE - Enschede, The Netherlands.
interior of the state space than based on the boundary behaviour as has been proposed by Gertsbakh [4].
The present paper investigates the shortest queue problem with threshold jockeying forc
parallel queues. The results in this case are even more striking, since, against the belief that the matrix-geometric approach provides any insight in the case c
>
2 (cf. Zhao and Grassmann [10]), we show that, by using the right criterion for selecting a partitioning, two partitionings can be constructed that help in solving the equilibrium equations completely. Actually, this paper has two goals. In the first place it presents a simple way of showing; that the shortest queue problem with threshold jockeying forc
parallel queues has a nice and elegant solution.In the second place it demonstrates that the use of a proper criterion for selecting a partitioning of the state space can be crucial for the success of the matrix-geometric approach.Consider a queueing system consisting ofcparallel servers. The service rate of serveriis
'Yj,
i=
1, ...,c,
where, for simplicity of notation, it is supposed that'Yl
+ ... +'Yc
=
c.
Jobs arrive according to a Poisson stream with rate cp and generate an exponentially distributed worldoad with unit mean. An arriving job joins the shortest queue. Incase of multiple shortest queues, one of these queues is selected according to some arbitrary probability distribution. Ifthe maxi-mal difference between the lengths of the c queues exceeds some threshold value T, then one job switches from the longest to the shortest queue. In case of multiple longest queues, the queue loosing a job is selected according to some arbitrary probability distribution. This system can be represented by a continuous time Markov process whose state space S consists of the vectors(n1, n2, ..., nc ) wherenj is the length of queuei, i= 1, ...,c. Due to the thresholdjock-eying the state space S is restricted to those vectors for which
I
nj - njI
~T for alliand j. Forc=
2 this model has been analyzed by Gertsbakh [4] and by Adan, Wessels and Zijm [1]. For arbitrary cZhao and Grassmann [10] use the concept of modified lumpability of con-tinuous time Markov chains to find the equilibrium distribution of the queue lengths. The spe-cial case ofinstantaneousjockeying(T=
1) has been analyzed by Haight [6] for c=
2 and by Disney and Mitchell [2], Elsayed and Bastani [3], Kao and Lin [7] and Grassmann and Zhao [5] for arbitrary c. .In this paper it is shown that the matrix-geometric approach developed by Neuts [8] is very well suited to analyse this problem.Insection 2 we show for a suitably chosen partitioning of the state space that an explicit ergodicity condition, which is obviously
P
<
1, can be derived from Neuts' mean drift condition. However, for that partitioning the associated R-matrix cannot be determined explicitly. Therefore, in section 3 we propose another partitioning for which the associated R-matrix can easily be determined explicitly. Actually, the latter choice generalizes the choice used in [1] for c=
2, which was suggested by a more direct way of solving the equili-brium equations. Gertsbakh [4] uses the matrix-geometric approach forc =2, but his choice for the partitioning does not lead to an explicit solution for the associated R-matrix.
-3-2. Necessary and sufficient ergodicity condition
Application of the matrix-geometric approach requires a partitioning of the state space. First define for 1=0, 1, sublevel Ias the set of states (nit ..., nc)eSsatisfying nl+...+nc
=
I. Then for each state(nlo , nc )at sublevelI>
(e-I)Tnone of the queues is empty and the tran-sition rates from this state are identical to the rates from the corresponding state (nl+l, ..., nc+l) at sublevelI+c. This suggests to define for all I=
0, 1, ... level Ias the union of the sublevels Ie, le+l, ..., le+c-l and to partition the state space S according to these levels withI =T, T+1, ... and to put the levels 0, 1, ...,T-1 with less regularbeha~iourinto one set The states at level I are ordered by sublevel, states from each sublevel being ordered lexico-graphically. For this partitioning the generatorQis of the following fonn, where the first class corresponds to the group oflevels 0, ...,T-1,Boo BOl 0 0 0
BIO Al A o 0 0
Q= 0 A 2 Al A o 0
0 0 A 2 Al A o
Corresponding to the partitioning of level I~T into the sublevels Ie, le+l, ..., le+c-l, the square matrices A0,A1and A2 are of the fonn
Ao,o AO•I 0 0
A I•o AI,I A I•2 0
o
A2,I A2.2 A2,3o
(1)
The Markov processQis irreducible and, since two states at levels
>
Tcan reach each other via paths not passing through levelsST, the generator A o+A1+A2is also irreducible. Thus theorem 1.7.1 in Neuts' book [8] canre~dilybe applied. Specifically the Markov processQis ergodic if and only ifwhere e is the column vector of ones and 1t is the solution of
1t(A o +A l +A 2)=O, lee=l.
By partitioning 1t as (1to, ..., 1tc-l) and the column vector e as
e=[~O],
1;;-1(3)
corresponding to the form(1)ofAo, A1andA2'we get as a result that inequality(2)reducesto
and the equations(3)to
1ti-lAi-l
. .
j+
1tjA i i+
1ti+lAi+l.
j =0, i=1, ..0' c-2 ,1tc-2Ac-2.c-l
+
1tc-lAc-l.c-l+
1toAO.c-l =0 .(4)
(5)
Since the flow from a state at sublevel1
>
(c-1)Tto sublevel 1+1 is cp (an arrival) and the flowtosublevel1-1 is c (a service completion) it follows that
AO.lel =cpeo, Ao.oeo=-e(p
+
l)eo, AO.c-lec-l = ceo ,Ai,i+lej+l =Cpei, Ai,iej= -e(p
+
l)ei' Ai,i-l ei-l= cei' i= 1, "0'c-2 ,Ac-l,oeo = cpec-l' Ac-l,c-l ec-l = -e(p
+
l)ec-l' Ac-l,c-2ec-2= cec-l 0Hence inequality (4) simplifiesto CP1tc-l ec-l
<
c1toeoand multiplying the set of equations(5)with ei leads to CP1tc-l ec-' - p(c
+
1)1toeo+
C1tl el =0 ,CP1tj_l ej-l - p(c
+
1)1tjei+
C1tj+l ej+l =0, i= 1, ...,c-2 ,(6)
CP1tc-2ec-2 - p(c
+
l)1tc-l ec-l+
c1toeo =00By the symmetry of these equations it follows that 1toeo = ... = 1tc-l ec-l and thus from (6) we can conclude that:
Theorem1: The Markov process
Q
is ergodicif
andonlyifp<
1.So, for the chosen partitioning of the state space, the mean drift condition(2)easily leads to the desired ergodicity condition, but the associated R-matrix, however, cannot be determined
-5-explicitly. Inthe next section we adapt the definition of the levels and show for this new parti-tioning that the associated R-matrix can be determined explicitly.
3. Explicit determination of R
We now adapt the definition of level I as the set of states (n1t ••• , nc)eS satisfying
max(nit ...,nc)
=
I. The state space S is partitioned into the sequence oflevels T, T+l, ....Thelevels 0, 1, ...,T-1 with less regular behaviour are put together in one set The states at each level are ordered lexicographically. For this partitioning the generatorQis of~eform
Coo COl 0 0 0 CIO Cn Do 0 0
Q=
0 Dz DI Do 00 0 D z D1 Do
The square matricesDo,D1andD
z
are of dimensionsm xm
wherem
is the number of states at a level ~T. The Markov processQ is irreducible and, since two states at levels>
Tcan reach each other via paths not. passing through levels:$; T, the generator Do
+D1+Dz is alsoirreduci-ble. Thus theorem 1.7.1 in Neuts' book [8] can again be applied. By partitioning the equilibrium probability vector
P
into the vector(Po, ..., Pr
-1)and into the sequence of vectors PT,Pr
+1, ••.where PI is the equilibrium probability vector of levell, we then obtain
PI=PrRI-r , I> T ,
where the matrix R is the minimal nonnegative solution of the matrix quadratic equation
Do +RD1+RzD z=0. (7)
(8) The matrixR can be determined explicitly due to the special matrix structure of Do. Since it is only possible to jump from level Ito level1+1 via state (I, I, ...,I), it follows thatall rows of
Do are zero, except for the last row. ThusDo is of the form
Do
=
[~
]
where v
=
(vo •...•
Vm-I) .Since rows in R which correspond to zero rows in Do, are also zero (see e.g. the proof of theorem 1.3.4 in [8]), we conclude that R is also of the form
R
=
[~]
wherew=
(wo ...Wm-l).Hence the matrix-geometric solution simplifies to
(9)
LetV, be the set of states(nl •...• nc)eSsatisfyingnl+...+nc=1. Then the componentWm -] is
detennined by balancing the flow betweenV, andV,+ltyielding forI> (c-l)T
P(V'+l)c =P(V,)CP.
and by applying this relation
c
times.P(V'+c)
=
pC
P(V,) .On the other hand. the setVcl is a subset of the union of the levelsI. 1
+
1. ...•1+T-1.so itfol-lows from (8) that forI
>
TP(Vcl)=Kw~_] •
for some constantKbeing independent ofI.Combining(9)and(to)yields
(10)
The remaining components of w are solved from equation(7).which. by insertion of the special fonns of R and Do. simplifies to
v+w(DI +Wm_ID 2)=0.
Finally. substitutingWm-I
=
pC
leads to w=-v(D] +pcD2)-I.where the inverse ofD]
+
pC
D2 exists. since this matrix can be interpreted as a transient gen-erator (escape is possible at least from the last state).Theorem 2: R=
[~]
where w=-v(D t+
P'D2)-t.Remark:
The explicit solution of R is mainly due to the special matrix structure of Do.Infact, Ramaswami and Latouche [9] show that if the generatorQcan be partitioned as in the begin-ning of this section andDo is given byDo=
x'ywherexis a column vector andyis a row vec-tor. then R is explicitly detennined. once its maximal eigenvalue is known.4. Conclusion
Inthis paper we studied the shortest queue problem withcservers and threshold jockeying and we showed that the matrix-geometric approach is very well suited to analyse this problem. The interesting point was that a proper choice for the state space partitioning depends on where one is interested in. One partitioning leads to an explicit ergodicity condition and for another partitioning the associated R-matrix is detennined explicitly. Another conclusion may be that. for the application of the matrix-geometric approach for solving the equilibrium equations of markov processes. the aspect of selecting a proper partitioning is of crucial importance. Inthe
-7-treated case the extension of gertsbakh's partitioning to
c
>
2 does not provide any insight. However, two other partitionings, based on the process behaviour in internal points, lead to a complete solution of the problem.References
1. ADAN, I.J.B.F., WESSELS, J., AND ZIJM, W.H.M., "Analysis of the asymmetric shortest queue problem with threshold jockeying,"Stochastic Models, vol. 7, 1991 (to appear).
2. DISNEY, R.L. AND MITcHELL, W.E., "A solution for queues with instantaneous jockeying and other customer selection rules,"Naval Res. Log., vol. 17, pp. 315-325, 1971.
3. ELSAYED, E.A. AND BASTANI, A., "General solutions of the jockeying problem,"Eur.J.
Oper. Res., vol. 22, pp. 387-396, 1985.
4. GERTSBAKH, I., "The shorter queue problem: A numerical study using the matrix-geometric solution,"Eur.J.Oper. Res., vol. IS, pp. 374-381, 1984.
5. GRASSMANN, W.K. AND ZHAO, Y., "The shortest queue model with jockeying," Naval Res. Log., vol. 37, pp. 773-787,1990.
6. HAIGHT, F.A., "Two queues in parallel,"Biometrica, vol. 45, pp. 401-410, 1958.
7. KAo,E.P.C. AND LIN, C., "A matrix-geometric solution of the jockeying problem,"Eur.
J.Oper. Res., vol. 44, pp. 67-74, 1990.
8. NEUTS, M.P.,Matrix-geometric solutions in stochastic models, Johns Hopkins University
Press, Baltimore, 1981.
9. RAMASWAMI,V. AND LATOUCHE, G., "A general class of Markov processes with explicit matrix-geometric solutions,"OR Spectrum, vol. 8, pp. 209-218, 1986.
10. ZHAO, Y. AND GRASSMANN, W.K., "SolVing a parallel queueing model by using modified lumpability," Research paper, Queen's University, Dep. of Math. and Stat, 1991.
P.O. Box 513
5600 MB Eindhoven, The Netherlands Secretariate: Dommelbuilding 0.03 Telephone : 040-473130 -List of COSOR-memoranda - 1991 Number 91-01 91-02 91-03 91-04 91-05 91-06 91-07 91-08 91-09 91-10 91-11 Month January January January January February March March April May May May Author
M.W.I. van Kraaij W.Z. Venema
J. Wessels
M.W.I. van Kraaij W.Z. Venema
J. Wessels
M.W.P. Savelsbergh
M.W.I. van Kraaij
G.L. Nemhauser M.W.P. Savelsbergh R.J.G. Wilms F. Coolen R. Dekker A. Smit P.J. Zwietering E.H.L. Aarts J. Wessels P.J. Zwietering E.H.L. Aarts J. Wessels P.J. Zwietering E.H.L. Aarts J. Wessels F. Coolen The construction of a strategy for manpower planning problems.
Support for problem formu-lation and evaluation in manpower planning problems.
The vehicle routing problem with time windows:
minimi-zing route duration.
Some considerations concerning the problem interpreter of the new manpower planning system formasy.
A cutting plane algorithm for the single machine scheduling problem with release times.
Properties of Fourier-Stieltjes sequences of distribution with support in [0,1).
Analysis of a two-phase inspection model with competing risks.
The Design and Complexity of Exact Multi-Layered Perceptrons.
The Classification Capabi-lities of Exact
Two-Layered Peceptrons.
Sorting With A Neural Net.
On some misconceptions about subjective probabili-ty and Bayesian inference.
COSOR-MEMORANDA (2) 91-12 91-13 91-14 91-15 91-16 91-17 91-18 91-19 91-20 91-21 91-22 91-23 May May June July July August August August September September September September
P. van der Laan
1.J. B. F. Adan G.J. van Houtum J. Wessels W.H.M. Zijm J. Korst E. Aarts J.K. Lenstra J. Wessels P.J. Zwietering M.J.A.L. van Kraaij E.H.L. Aarts J. Wessels P. Deheuvels J.H.J. Einmahl M.W.P. Savelsbergh G.C. Sigismondi G.L. Nemhauser M.W.P. Savelsbergh G.C. Sigismondi G.L. Nemhauser
P. van der Laan
P. van der Laan
E. Levner
A.S. Nemirovsky
R.J.M. Vaessens E.H.L. Aarts J.H. van Lint
P. van der Laan
Two-stage selection
procedures with attention to screening.
A compensation procedure for multiprogramming queues.
Periodic assignment and graph colouring.
Neural Networks and Production Planning.
Approximations and Two-Sample Tests Based on P - P and Q - Q Plots of the Kaplan-Meier Estima-tors of Lifetime Distri-butions.
Functional description of MINTO, a Mixed INTeger Optimizer.
MINTO, a Mixed INTeger Optimizer.
The efficiency of subset selection of an almost best treatment.
Subset selection for an -best population: efficiency results.
A network flow algorithm for just-in-time project scheduling.
Genetic Algorithms in Coding Theory - A Table for A] (n, d) .
Distribution theory for selection from logistic populations.
J. Wessels W.H.M. Zijm
of the shortest queue problem with threshold