A Proof for the Queuing Formula: L= #W

John D. C. Little

Operations Research, Vol. 9, No. 3. (May - Jun., 1961), pp. 383-387.

Stable URL:
http://links.jstor.org/sici?sici=0030-364X%28196105%2F06%299%3A3%3C383%3AAPFTQF%3E2.0.CO%3B2-I

Operations Research is currently published by INFORMS.

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained
prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in
the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/journals/informs.html.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

The JSTOR Archive is a trusted digital repository providing for long-term preservation and access to leading academic
journals and scholarly literature from around the world. The Archive is supported by libraries, scholarly societies, publishers,
and foundations. It is an initiative of JSTOR, a not-for-profit organization with a mission to help the scholarly community take
advantage of advances in technology. For more information regarding JSTOR, please contact support@jstor.org.

http://www.jstor.org
Fri Feb 29 10:10:44 2008
 A PROOF FOR THE QUEUING FORMULA: L = XW

John D. C. Little
Case Itlstitc~teof Technology, Cleveland, Ohio *
(Received November 9, 1960)

In a queuing process, let I/&be the mean time between the arrivals of two
consecutive units, L be the mean number of units in the system, and 1Y be
the mean time spent by a unit in the system. I t is shown that, if the three
means are finite and the corresponding stochastic processes strictly station-
ary, and, if the arrival process is metrically transitive with nonzero mean,
then L=XIV.

H EURISTIC arguments are sometimes given lo show that, in a steady-

state queuing process, the following formula holds:

whcrc L =expected number of units in the system

T17=expected time spent b y a unit in the systeln
1 / X =expected time between two consecutive arrivals to the system.

Expression (I) is of interest because it is sometimes easier to find L than

W (or vice versa) in solving a queuing model.
A brief plausibility argument for rather general validity of (I) is given
by MORSE(reference 1, p. 22). He goes on to prove it in a number of spe-
cific models. GAI,I,IHE~~[~] estahlishes it for the case of Poisson arrivals
which have a rate independent of queue length and which come to a multi-
ple channel facility having qfirst-come, first-served discipline. Wr shall
prove it under assumptions considerably more general.
By a queuing process will be meant a mathematically specified opcl.at,iou
in which units arrive, wait, and then leave. I t is presumed that t,he opera-
tion thereby generates three well-defined stochastic processes:
(nt, - m < t < ) =the number of units in the system a t time t
{w,,- < r < ] =the time spent in the system by the rth arriving unit
{TT,- <T < 1 =the time between the arrivals of the rth and (r f1)st units
to the system.

These processes are defined on some space Q and any point o E Q selects a
* This work was supported in part by the Air llevelopment Center, Wright-
Patterson Air Force Rase, United States Air Force, under Contract No. AF33(616)-
6446,
303
384 john D. C. Cittlc

function and two sequences,

which represent a specific realization of the queuing operation over all

time. The random variables nt, w,, and T, are nonnegative.
The time of arrival of the rth unit will be denoted t, and is defined by

For convenience we choosc

The following relation is taken to be part of the definition of a queuing

process: Let
1 for x 2 0 ,
u(x) =
0 for x<O;
then, for any w, n,= C?:u(t-tj) u(tj+wj-t). (2)
This relation says that the number in the system a t t is the number of
units whose time of arrival is before (or equal to) t and time of departure
is after (or equal to) t.
THEOREM 1 : If, in a queuing process, ( i ) each of the stochastic processes
nt, w,, and Tr is strictly stationary with finite mean, and (ii) the rTprocess is
metrically transitive with mean T=l/X >0, and, if we let

L(w) =lim
t-m
l
-
t
1 t
. ( a ) ds,
1 "
W(w) =lim -
m+m m 1
wj(w),
(3)
1 "
T(w) =lim
nb-m
-
m
CT~(w),
1

then, with probability 1, the limits in (3) exist, are finite, and satisfy
W(w) = T(w) L(w). (4)
The existence and finiteness of the limits is an immediate consequence
of the ergodic theorems for strictly stationary stochastic processes (see
DOOB,reference 3, pp. 465 and 515).
Consider a specific point wEQ. Let t, denote the length of the inter-
val [0, tn,(w)). Define
 Queuing h r r n z t l n 385

I n order to take the limits of ( 5 ) sin~ultaneously,me first show that as

m+w, t,+w w. p. 1 (with probability one). By the ergodic theorem,
the metric transitivity of thc 7,. process, and its nonarro mean, we have
l/T,(w)+I/T(w) = 1 / T < cc w, p. I . Let a = ~ ( w-tl(w). ) We scr
that O<a< cc -cv. p. 1. Thcn l/l',(w) =m/(t,+a)+limm/t, w. p. 1;
limmlt, < a, w. p. 1; so that m+ cc implies t,+ a, w. p. 1.

Fig. 1. P a r t of a specific queuing realization w , showing the number in the sys-

tem a t time s , n,; the wait in the syst,em for the 7th arrival, w,; and the interarrival
time started by the rth arrival, rr. (The figure is tlmwn for the case of departure
in ortler of arrival, but this is not required for the proofs in the text.)

1ntegrat.ing (2) for fixed w gives

where v(x) = x for z >0 and v (x) = 0 for 2 5 0. The situation is illustrated
ill Fig. 1. The area under the curve n, from 0 to t, is, except for certain
carry-over effects a t the ends of t,he interval, the sum of the waiting times
of the units that arrived during the iiit,erval. These carry-over effects are
indicated by the areas A and B, which correspond to the last two sums on
the right in (6).
Dividing by m and using (5) gives

The last two terms oil the right can br shown to go to zero w.p. 1 as
386 John D. C. Little

m-+m : In the last term the sum consists of a finite number (no) of finite
terms except on the union of (lao+ 1 ) w-sets of probability zero. Thus the
sum is finite w, p. 1, and, since it is iadependcnt of m, the desired limit is
zero w. p. 1. Ill the next to last term, L,n-+L(w)< ca and a/m-+O w. p. 1.
Thus

If now we consider the intcrval (t-.,,,(a), 01 and define L-,, H7-,,, and
T - , n ~ ~ ~ l o g ~to
u stheir
l y cou~~terpnrts
above, e.g.,

the11 the symmetry of the ergodic theorems with respect to time aild argu-
ments the same as used previously yield
TV(w) - l'(w)L(w) = -lim(l/m) x,s-,z~(zo,+t,-t-,) 5 0 m,p. 1.
Therefore, TP(w) = T(w)L(w) w,p. 1
as was to be shown.
THEOREM
2: Let

then, uncl~rthe hypotheses of 'I'heurem I ,

W = TI,.
The ergodic theorems state that for almost a11 w the limits (3) are the
conditional expectations:
L(w) = E(nolg,], Tq7(w)= E{wol~b), T(w) = E{7018~]
where g,, ga, aiid 9, are the Bore1 fields of invariant subsets for the corre-
sponding processes, Since the T? process is metrically t,ransitive,
Y(w)= T,
and (4) hecomes Il'iw) - Tl,(w) w. 73. I.
Intcgratior~over il givr5, hy tlrfj~~iiiol~
of col~dilioilalexpectation,
TiT= T'l,
as was to be shown.
DISCUSSION

THEOREM 2 is the priiicipal result for applicatioiis and shows that ( I ) is a

valid relation among phase averages. Theorem I , oil the other hand, is
 Queuing Formula 387

perhaps more basic for it shows that an equivalent of (I) using time aver-
ages holds with probability one for any specific realizatioii of the queuing
process.
The results are remarkably free of specific assumptioiis about arrival
and service distributioiis, independence of interarrival times, iiumber of
chaniiels, queue discipline, etc. A requirement is made for strict station-
arity (although this is probably iiot the weakest requirement possible), but
the steady state iii most current queuing models mould appear to be strictly
stationary. Similarly, iii cases of practical interest, the arrival process is
likely to be metrically traiisitive.
Notice that the definition of what constitutes the 'system' is left flex-
ible. In conventional usage, the number of units in the system refers to
the number in queue plus those iii service. The theorem here, however,
only requires coiisisteiicy of meaning in the phrases, 'number of uiiits in
the system,' 'time spent in the system,' and 'arrival to the system.' Thus,
if we choose to label the queue as the system and let L, and W, refer to the
mean number and meail wait iii queue, me obtain

Similarly, if we have a model with priority classes i = 1, 2, . . . , p, and let

L, be the mean number of priority i units present, 7 i 7 , the mean wait of a
priority i unit, and 1 / X , the mean interarrival time for priority i units,
then
L,= A, TB,.
Morse (reference 1, p. 75) asks when ( I ) does not hold. As ail cx-
ample, me cite a type of model, used in his book and else~vhere,in which
arrivals come with rate X hut not all arrivalq join the system. Then ( I )
does not hold. However, inspection of the tlieorcm shows that ( 1 ) mill
hold if X is redefined to iiic1:dc only thosc arrivals that join the system.
iilteriiatively, we can say that ihc uiiith that do iiot join have a zero wait-
iiig time in the system aiid include thcrn in the calculatioil of TV. This
too will make (1) hold.
THE AUTHOR thanks DR. H. SLTVTOA GARBLK.for several constructive
suggestions about the proof. Tn particular, equation (2) and its explicit
definitioi~alimplicatioi~sare hi\.

KEFEKEKCES
1. P. M. MORSE,Queues, Inventories and Jfaintenance, TTriley, New York, 1958.
Notes on O p ~ ~ o t iResearell
2. H. P. GALLIHER, o ? ~ ~ I,Y;N, Chap. 4, Technology Press,
Cambridge, 1959.
3. J. L. Doou, Xtochastic Proc~ss(s,\Tile;, Xc\v STorb, 1953.