You are on page 1of 2

Chapter 7 Problems 249

9. Lavenburg, S., and M. Reiser "Mean Value Analysis of Closed Multichain Queueing
Networks," Journal of.the ACM vol. 27, no. 2, April 1980.
10. Perros, H., Queueing Networks with Bkicking. Oxford, New York, 1994.
11. Solberg, J., "Capacity Planning With a Stochastic Workftow Model," AIIE Transactions,
Special Issue on Current Research Directions in Production Planning and Control, 1980.
12. Solberg, J., "A Mathematical Model of Computerized Manufacturing Systems," Pro-
ceedings, 4th International Conference on Production Research, Tokyo, August, 1977.
(Also reprinted in Production and Industrial Systems, eds. R. Muramatzu and N. A.
Dudley, Taylor & Francis, London, 1978, pp. 1265-1275.)
13 R Suri G. W.W. Diehl, S. D. Treville, and M. J. Tomsicek, "From CAN-Q to MPX:
• E~oluti~n of Queueing Software for Manufacturing," Inteifaces vol. 25, no. 5 (19 )
128-150. 95
14. W. Whitt, "The Queueing Network Analyzer,
vol. 62, no. 9 (1983) 2779-2814. ,, The Bell System Technical Journal

Chapter 7 Problems

I The first few problemsfcomeupes


are alternative designs for
Besides illustrating the of the parameter values the same,?what is the expected total time
· • 1 tworks o qu · · to pass through the system now.
some very s1mp e ne rinciples of network design;_ ~ou
theory, they expose some p ative results as well as obtammg
should take note of the compar h e a Poisson arrival process
them. To start, assume that we . a\e That stream feeds two
with A= 0.9 customers per rmnu h. t has µ, = 1.0 customers
. eac h ith. a server
queues in senes, 7 8t a
per minute, as shown m Figure . .

--c~~
I FIG U RE 7 . 8 A simple open network
FIGURE 7.10 A modified network

t values t e sarn '


of the pararne er .
to pass through this van
fi t two queues can also be
Now, suppose that the rs_ Fi ure 7.11. Keepin~ all
;erged, giving the systehm sho;:~:t is~he expected total time
d . ?
.ant of the network es1gn.

a. What is the expected tota l fme


1 to pass through this
system? . rocess has A = 1: 8•
b. Now imagine that the Poisson amvad1pmly assigned (wi th
. . customers are ran ollel subsystems, each
but the arnvmg .
probability 1/2) to either of two para ld be as shown
10 11 A further modified network
like the one above. The network w~:me to pass through
Figure 7.9. What is the expected tota F I G URE 7 . . . times to what would ~suit
this system? . compare these wa1tm~ sin le station with a
4. Fm~~y:Uerge all four servers mtol: ord!r to have compa-
if you cou a simple M/M/4 syste~.e n·me would have to be
__JO d queue- an serv1c (Thi
I share ter values, the me . ns so use µ, = 0. . s
rable param_e for the two-stage vers: 'ry in this chapter, but it 5
double t_he n:;s not require any o_f th~s:~ way. Be sure to take
letes t e . esults from e
r
calculation dh comparison in an mte th our different ways of
)
1 I comp
t of the comparative ffi rtream w1'th the same resourcr,S, , .
F GU R E 7 . g A larger network no e_ the same tra c s . ckle a more realistic design
servm\he next few quest10ns ~a explored the formulation ofa
5. . In Chapter. model2,of a casealstlu y ter. In that model, time was
2S~b stNext, suppose that the final qu~u~i
s for Ofthe parallel situation. a c cen
ureeach
7 .1 0. 'Keepmg
· all Markov cham
sy ems can be merged, as shown 10 g

255
J
250 Chapter 7 Networ ks of Queues

questions (molds) was eight. Assume that the s .


anYable to use ers . (Th erv ce .
represented by transfers; we could not 4answer we were tations is six mmutes. e numbers are not 1 ·''nicat~
about how long calls lasted. In Chapter ~rate real time and s ke the calculations relatively clean so th realistic. but 1\
ma , at You c· \,i
rewards to extend the model to mto et represent contention by hand.) ijn do~~. I
operator costs; however, we could no_ y t"me that would result. a. The goal was to keep the machines Work' I
1
for resources and the add.iuona
· I queuemg 1· . ode I of a comp 1ete Possible, so calculate the utilizations of thing ai lliuc1 I
Now, we have the tools to build a r;a •s:t:s of each type, and station I. e lllach·1~ 1iji I
call center, including the number O . op~ t rms of the expected
we can evaluate the quality of serV1c_e m e we want to focus on
delays that customers encounter. In this case, h" h are W ' s.
b One way to increase the utilization is to add
• the system. What would the result be if you mored11101di~ I
the times spent wmtmg . . to spea k to. an operator,
.
The stations will be: I = Sales, with SIX operators,
w .1c2 = Billing,
q
d _ a
more? (Hint: Just add a row to the table v a ~ IJii;
calculate G.) · ,au Uleij
I
with four operators; 3 = Support, wi~ fouroper~tors; an 4 - at
single supervisor. Calls arrive according to a P01sso? process h 10. Conti~uing the problem 9 above, imagine that 11,: I
ilie rate of fifty per hour. The routing is as descnbed by e management 1s concerned about how the productivit
Markov chain developed in Chapter 2's Section 2.8. The service the shop is affected by the ~bsence of a ~orker. What :llil: j
times are as described in Chapter 4's problem 7, page 138. the utilization of the machmes become 1f one of the thr~
workers were absent?
a. Find the effective arrival rates to the four operator stations.
(Hint: The external arrivals are dividing into three streams 11. To gain a better understanding of how to use aclOSoJ
by the first state in the Markov chain model for transfers.) network with a dummy station to represent (approximately)i!I
b. Find the expected delays (Wq's) at each station. (Hint: The open network, consider a very small and simple example.F~
formulas for the Lq's are given in Chapter 6, Table 6.3, the open network, use an ordinary MIMI I queue with J. =O.!
page 205. Remember to use the correct definition of p for and µ = I. All of the formulas in Section 6.7 will apply. To
each model.) create an approximately equivalent closed network, we _~,11
6. Extending the problem 5 above, suppose that one of the need a dummy station that acts like the arrival process. That1s.11
delay times is simply too long to be acceptable to customers, so will have one server that completes service at a rate that1sequal
another operator is to be hired. First figure out which type of to the arrival rate of the open network. We must also hm
operator is most needed, and then recalculate the delay times enough customers • crrcu
. 1atmg
. • th e sys tern to ensure 1haL
m
after that change is made. almost always, the dummy server can contmue · tO work· When· Id lt
7. Although you will want to use software (such as the code in ever it goes idle, the simulated arrival process woud·u·oo
Appendix mto compute the normalization constant for problems of .
rnterrupted. We can measure how close we are to that. con. ,·o1n of
realistic size, you should work a few small problems by hand in order (and hence how good the approx1mat1on. . is• ) by the ull_ 11za1
JOforthl
to understand what the algorithm does. An easy one to start with is the dummy server. So let us start with a value o'.N - a]relatiw
the one shown in Figure 7.2. In order to keep the hand calculations number of customers in the closed system. The intern for thr
simple, use workloads of two for each station and N == 4. arrival rates in the closed network will be ~qual 5~ I mean i;~e
a. Find the normalization constant G, using the table method
described in Section 7.9.
b. U~i?g _the table and the results in Table 7.2, find the
common value. The service rate (the reciproca •valratevf
service time) of the dummy server must match;e
the open network, so use f> = 1.25 or 5/4 for
::my.~
but you ni3)
uht1hzat1on of both _servers (which should be equal, because numbers are clean enough to work with by hand.' us)
t ey are symmetric). · to 0 te toserver
want to use a spreadsheet if the work is · wh(il
8.
t Repeat the same problem using two stations th t h a. What is the utilization of the durnrnY
wo servers eac h.. Agam,
. the utt.11zations
. should come outaequal
ave
but the calculations and results will be different. ' -
N- 10?
.
1 erver in
·,j11JI

9. Problem 15 in Chapter 5 had you model .


b. Compare the utilization of the rea s •..,arioH- .1
nu1vu I and in the closed networ a
l\.,f"' .. , k pprox1a... [he(J(lg . io~
. Markov process mvolvmg
time . . recycling molds a con11nuous
d•
" . process. At that time,
. 1server 111
u,actunng it was treated as useb. m
h man - c. Compare the values of L for the rea ·marioil•
process. However, it could also be treated as a Irt -death MIMI I and in the closed network approx• the jl!ll~
network of queues. Th e fi rst stat10n
. (machines) h· da two-statio n
~ontinuing the problem 11 abOve rer approX11·11au
nP.81 "(111'
12•
the secon d station
· (workers) had three. The numbd twof ·servers,• calculations with N = 20 to see how th
ero custom- improve.

You might also like