You are on page 1of 1

systems

Single server stochastic recirculation

gl

second term represents the contribution from the process


representing leg 2 atthe previous cycle.

rder

th

we

interested reader to Whitt


[2,3] for details.
An overview of the composite approximation technique is provided
in Fig. 2. we start in box I by
approximating the stationary counting process aloni the top
leg by
interweirt distribution has the same first two moments as the countiog "-r"n"*"i fro."r. ,"t or"
pro""r..
to either

B;#il;

box 2A (cr < 1) or box 29 (cq,) 1), we analyze the behavior ut ti. q""u"ine
,;liri". *irre
Proposition l- we should note that given the results of box l,
there is no further-upprJ*i*ution

ln
boxes 2A or 28. we finish-the-cycle by approximatiog tn
rup"6.ition of the
with
the primary input in box 3, after which we returnto box l.
In our tests of the method we started the
system with no units in the system, and went through a number
of cycles until a steady state situation
was achieved' The decision node before box 2 is ni*rary
because iuring uny on.'"ipri"uiion ortn"
algorithm it is possible to use both boxes a number of diiferent times
on the way to steady state. An
algorithm based on Fig. 2 was coded, and run for various
service and primary inpui rates and
coeffrcients of variation. A comparison of the algorithm with
a simulation is discussed in Section 5_

"rJ;;;;;"ess

4. SIMULATION RESULTS

In this section we compare the results of our approximation algorithm with


a simulation of the
stochastic recirculation system for various values oiinput and
servic-e parameters. A summary of both
the simulation and algorithm resurts for these various parameter
3. The
algoritbm results were oltlinea br starting with un
ry.tem,and applying the algorithm until a
steady-state was reached. The simulation iesults were
".pty
obtained by averaging ten different runs, each.
run lasting until the steady state was achieved. The numbers in
iarentties"I ."p."r.niii" ,t"roura
deviations of the ten different runs.
consider Fig' 3' Comparisons can be drawn between the rates and
coeflicients of variation along
the topandbottom legs of the systan for low (p:0.3), medium (p :0.S)
and high (p:0.7) traffic
intensities. Two types
service disciprine.were considered: exponential (c5:l) and generar
_of
hYnelelnongntial (cr:1.5). comparingihe simulation *ittr
ttr afiroximation shows an excellent
match for the rates 2r, ,l'r. Results for the coellicients or
c1, c2, although perhaps less
important to the design and operation of the system, are also quite
satisfactory. All the steady-state
results display equilibrium between the rate of overflows
from the server and the rate of superposition

r"I;;;;;;iil

|if;

uu"tiio

ofprocessesatthesourceofprimaryinput.considertheparameteivatra
rate of

i,

:3'19

along the top leg, along with the appro^xim"r"a

1l
0.3
0.3
0.7
0.3
0.5
,O.7
0.3
0.5
o.7
0.3
0.5
o.7
0.3
0.5
o.7
0.3
0.5
0.7

0.5
0.5
0.5

1-

'l

1.5
1.5
.1.5

0.33
0.82
2.5s

o.u

,l
1
1

0.5
0.5
0.5

't.5

.l

1.5
1.5

1.5

1.5
1.5
1.5

1.5
1.5

1.5
1.5

1.5

li
0.34(0.00)
0.86(0.02)
2.50(0.08)
0.45(0.01)
1.15(0.03)
3.32(0.15)

1.11
3.19
0.49 0.4S(0.01)
1.31 1.25(0.03)
4.11 3.41(0,16)
0.35 0.36(0.00)
0.91 0.90(0.00)
2.61 2.61 (0.11)

o.M

0.44(0.01)
1.16(0.04)
3.25(0.18)

1.11
3.11
0.48 0.48(0.01)
1.28 1.26(0.06)
3.8s 3.s1(0.26)

cl*

crt

0.57 0.58(0.00)
0.77 0.77(0.01)
1.13 1.07(0.02)
1.08 1.07(0.10)

"r,

r,=0.7,co:),6:cr-l.A

yi"ras an overflow stream with a

It

lzt

^2

C2'

czt

1.10
1.24

1.01 (0.01)

1.46

1.(0.04)

0.03 0.05(0.00) 1.@ 1.00(0.08)


0.32 0.36(0.02) 1.06 r.17(0.03)
1.85 .81 (0.08) 1.29 1.38(0.00)
0.14 0.15(O.Ol ) 1.22 1.28(0.02)
0.61 0.65(0.03) 1.35 1.46(0.04)
2.49 2.63(0.15) 1.55 1.6s(0.05)
0.19 0.19(0.01) 1.41 1.46(0.02)
0.81 0.76(0.02) 1.60 1.64(0.0s)
3.41 2.72(0.15) 1.88 1.81 (0.03)
0.05 0.06(0.00) 1.20 1.22(0.05)
0.41 0.43(0.09) 1.21 1.36(0.03)
1.91 1.92(0.11) 'r.34 1.50(0.04)
0.14
0.15(0.01)
1.29 1.35(0.03)
0.61
0.67(0.03)
1.40 1.55(0.03)
2.41
2.58(0.20)
1.55 1.74(0.07)

't.49

1.3s(0.03)
1./t3(0.02)
1.s7(0.04)

0.18
0.78
3.15

1.2O

1.4

1.18(0.04)

1.'O(0.11)

1.47 1.35(0.02)
1.56 1.39(0.03)
1.82 1.s4(0.93)
0.6s 0.61 (0.01)
0.89 0.80(0.01)
1

.17 1.19(0.27)

1.58

1.79

1.22(0.02)

0.18(0.01)
0.77(0.06)
2.81(0.25)

rAn approximation
outcome.
tA simulation outcome.
Fig. 3. Comparison of approximation and simulation ourcomes.

.47

1.63
1.8s

1.51 (0.03)
1 .71 (0.03)

1.86(0.04)

You might also like