international journal of

production
economics
ELSEVIER !nt. J. Production Economics 55 I 1998) I 13- ! 19

A quantitative approach to estimate the size of the time buffer in the

theory of constraints
Zinovy D. Radovilsky*
&'hool of Business and Economics, Cal(Ibrnia State Unirersio', Hayward, Hayward, CA 94542, USA
Received 19 August 1993; accepted 19 August 1997

Abstract
This paper describes a new approach in calculating the optimal size of the time buffer which is used to accommodate
disruptions in production processes and provide maximum productivity of capacity constrained resources (CCR). The
problem is formulated in terms of single-server finite queue. The new approach is based on identifying the optimal
number of units waiting in the line (the optimal size of the time buffer) that will maintain the highest operational profits
while protecting CCRs from becoming idle. Sensitivity analysis of the derived results can help to schedule production at
CCRs. ~(" 1998 Elsevier Science B.V. All rights reserved.

Keywords: Time buffer; The theory of constraints

I. Introduction The core of the theory [7] involves identifying

the system constraints and making decisions about
The theory of constraints (TOC), or synchronous how to work with these constraints to synchronize
manufacturing, has become a valuable system in a production process. System constraints can be
modern operations management. Using the ideas workers, machines, market demand, company pol-
and methods of the TOC, companies have achieved icies, and regulation affecting a company. We will
a large reduction of work-in-process and finished- consider only physical constraints, i.e. those asso-
goods inventories, significant improvement in ciated with resources involved in the production
scheduling performance, and substantial earnings process, such as workers, equipment, and materials.
increase [1-4]. Today, the TOC serves as a valu- A physical constraint is usually a bottleneck
able addition to or even as a substitute for such resource - that is, any resource whose capacity is
well-known manufacturing systems as material less than or equal to the demand placed upon it
requirements planning (MRP) and just-in-time [8, 9].
(JIT) [5, 6]. Recognizing capacity constrained resources
(CCRs) is essential in the TOC. A CCR is a resource
which, if not properly scheduled and managed, is
*Tel.: ( I ) 925 648 4895. ( 1) 510 885 4176: Fax: ( I ) 510 885 2 ! 65: likely to cause the actual flow of products through
E-mail: zradovil(a,csuhayward.edu. a plant to deviate from the planned product flow

0925-5273/98/$19.00 Copyright ~" 1998 Elsevier Science B.V. All rights reserved
Pll S 0 9 2 5 - 5 2 7 3 ( 9 7 ) 0 0 1 3 1 - X
114 Z.D. Radovilsk3,/int. J. Production Economics 55 (! 998) 113-119

[10]. Production output depends directly on how in the production process. But a time buffer that is
the system handles CCRs. Any disruption in a pro- too large may cause a significant waste of resources
duction flow caused by ineffective usage of a CCR by accumulating unnecessary inventory stock in
decreases production output. The main TOC front of the CCR.
technique for identifying and exploiting CCRs is The author of the Theory of Constraints, Eliyahu
referred to as drum-buffer-rope (DBR) [10, 11]. Goldratt [8], stresses that determining the time
A "drum" is a control point in the production buffer's optimum size is not an easy task. Literature
system associated with a CCR. The name "drum" is sources show different ways to approach this prob-
derived from the fact that this control point (CCR) lem. The majority of the authors relate the size of
will set the timing for the rest of the system. "Rope" the time buffer to the production lead time. Some of
is the term used for the communication feedback to them [10] recommend that the total size of the time
the resources before the CCR so that each of them buffers should be set at approximately half of the
produces only the amount of inw~r,~,,: ~h~t the manufacturing lead time. They have identified such
CCR can complete. size by experience without any specific quantitative
The central part of the DBR technique involves considerations. Others [12] estimate the size of the
inserting time buffers to protect the system from the time buffer as one-fourth of the total load time of
inevitable fluctuations of processing time. These the system. In these empirical approaches, the size
fluctuations are usually caused by the internal d~s- of the time buffer is viewed as a proportional part of
ruptions that occur in manufacturing and service the manufacturing production time, which is the
environments. The time buffer increases the pro- sum of processing, setup, transportation and queue
cessing time from the absolute minimum required times.
to process products by an amount of time sufficient One researcher [13] asserts that, since syn-
to accommodate these disruptions. Practically, this chronous manufacturing is a process of continuous
means that the process is planned with excess time improvement, there is no need to calculate the
in the schedule (a time buffer), which causes some initial size of the time buffer. This approach identi-
inventory build-up in front of the CCR to protect fies buffer points, then tries different-sized time
the resource from fluctuations in the production buffers during the actual production process, cor-
process. Contrary to the JIT system, in the TOC recting the sizes in the light of experience. The size
the time buffer is considered as a manufacturing of the buffer will decrease as improvements are
strategic reserve that can protect the system from made. However, if the original buffer size was
fluc'~uations in the process. chosen without preliminary evaluation, it may take
According to the principles of the TOC, the best a significant amount of time (even months) to de-
locations for a time buffer in a production flow are crease it to the appropriate level. But studying
in front of a CCR and after the final operation (final successful companies shows that the initial set-up,
resource) [10-12]. Placing a time buffer in front of as a starting point for improvements, is based on
a CCR means that the CCR will have enough time analysis and calculations.
to complete a required job without delaying the Approaches to identifying the size of the time
total production completion. The time buffer after buffer presented in sources are very empirical. They
the final resource will prevent delays in operations do not consider fluctuations of the production time
after CCRs. before the CCR, which is a critical issue in the
A number of sources cited describe identifying TOC. Also, such approaches do not analyze and
the CCRs and determining the location of the time compare the costs associated with different sizes of
buffers. Much less attention has been given to de- the time buffer against the benefits it produces.
termining the size of the time buffer. However, the Goldratt [8] pointed out that setting the time buf-
size of the time buffer affects the success of a process fer's size involves a trade-off between the value of
using the drum-buffer-rope technique. The smaller revenue, from one side, and the operating expenses
the time buffer, the greater the likelihood that it will and inventory costs associated with the time buffer
not protect the CCR from some of the disruptions from the other.
 ZD. Radovils~,/hn. J. Production Economk's 55 (1998) 113-!19 115

2. Modeling the size of the time buffer the results and studies of many researchers
[-14-17]. This assumption, a cornerstone of queu-
Let us consider a CCR that manufactures an ing theory, leads to relatively simple formulas for
order at a specific stage of the production process. the characteristics of the queuing system.
The CCR can be described as a server that In a M/M/1/K system [18], the probability that
processes incoming products. Because this server's a server will be idle is:
capacity is close or equal to actual production
demand, fluctuations in incoming flows can make l--p
this CCR into a bottleneck resource, with resulting Po - 1 - - pK + 1 (p -'/=l),
effects on the production process. (1)
To prevent such situations, the CCR is protected 1
by a time buffer, which can be described as a wait- P o - - (p = 1),
K+l
ing line of incoming products in front of the CCR.
The shorter the waiting line, the more likelihood where p is the utilization factor, and p = 21p.
that the CCR may become idle and delay the pro- According to Eqs. (1), this probability is a func-
cess. The longer the waiting line, the more operat- tion of the number of units in the waiting line. To
ing expenses are tied up in the production. Thus, protect the CCR from being idle, we need to have
the problem is to identify an optimal capacity of the a number K which will not allow Pe to exceed an
waiting line (optimal size of the time buffer) that appropriate limit x:
will maintain the highest profits in the operations
while still protecting the CCR from becoming idle Po(K) <~~. (2)
with some confidence level.
The description we presented perfectly matches Solving inequality (2) for K, we obtain the fol-
the formulation of the problem associated with lowing expressions:
queuing theory. From this perspective, an "'arrival
of a customer" simply means a new unit of incom- I n ( (p+~-I))-x-
ing product added to the time buffer. The average
K~> -1 (p~> 1),
number of units added to the time buffer per unit lnp
time is designated as 2. Service can be identified as (3)
an operation of utilizing products from the time 1
K>~--I (p = 1).
buffer with the average service rate of p units per
unit time. The number of units in the line represents
the size of the time buffer. The optimal size of the We consider 2 to be greater than or equal to p, or
time buffer can be described as the maximum re- p >f 1. If,;. is less than p, the capacity of the resource
quired number of units in front of the CCR. If the is greater than incoming demand, in which case
waiting line has more units, it becomes overloaded the resource is neither a CCR, nor a bottleneck
with inventory. In case of less units than the max- resource. Inequalities (3) give a helpful estimate
imum required, the time buffer may not protect the of K, and thus of the possible size of the time buffer.
CCR from possible disruptions in the system. These A manager who wants to apply the TOCs ideas,
necessitate to use a finite queuing system for identi- particularly the drum-buffer-rope approach, will be
fying the optimal size of the time buffer. building a time buffer on the solid ground of a sci-
Such a formulation allows us to describe the entific estimate.
mode! as M/M/1/K, a single-server system with Inequalities (3) allow us to identify an optimal
a finite queue of no more than K units, Poisson size of the time buffer. This optimization should be
distribution of arrivals, negative exponential distri- based on a trade-off between the benefits of using
bution of service time, and the queue discipline a particular size of the buffer and the expenses
"first come, first serve". The assumption of Poisson associated with handling inventory and supporting
arrivals and exponential service time concurs with that buffer (operating expenses). Net profit in the
116 Z.D. Radorilslo'/lnt. J. Production Economics 55 (1998) 113-119

TOC can be calculated as the values of Po and L~, respectively. The larger the
buffer size the lower the probability of the server
NP = TH - OE, (4) being idled. This leads to the larger throughput,
but, at the same time, increases the cost of carrying
where NP is the net profit, TH is the throughput,
inventory stock in front of the CCR due to the
OE is the operating expenses.
larger buffer size. As a result, by varying K, we can
Throughput (TH) is one of the most important
determine a particular (optimal) value of the buffer
financial measures in the TOC. It is defined there as
size that maximizes net profit.
a difference between overall return generated
When p = 1, the optimal value of K can be easily
through sales and material costs (inventories) of
identified by substituting L~ and Po in (8) with their
goods sold. Throughput is also described as the
expressions from (1) and (7); differentiating NP with
overall amount of income generated by the system
respect to K; and setting result to zero:
through sales. Assuming that throughput is con-
stant per unit of sale, we can present it in our model K K
as following: NP(K)=I~I +KCT" ~-CoE.

TH = lt(l - Po)Crr,. (5)

t~NP(K) lL 1
t~K = (1 + K) 2 CTH -- ~ COE -- 0. (9)
where Cm is the throughput per unit of sale
(a difference between unit price and unit cost of
materials). Solving the last equation for K, we receive the
The lower the value of Po, the larger the through- following:
put that the CCR can produce. (In such a formula-
tion we assume that all products completed by the K* / ~ICTH
= X/ COE l (p = l) (10)
CCR will be sold.)
Operating expenses (OE) in the TOC represent
Formula (10) is an EOQ-like formula. It is de-
a difference between total costs and material costs
rived from an assumption that the arrival rate is
of products. In terms of our particular formulation,
equal to the service rate, which means no excess
these expenses include a cost of carrying a stock
stock or shortages of inventory is allowed in front
associated with the time buffer:
of the CCR. This simultaneously represents one of
OE = LsCoF, (6) the assumptions of the classic EOQ model. A fact
that we determined the EOQ-like formula for
where L~ is the average number of units in the a situation when p = 1 proves a validity of our
system, COE is the carrying cost per unit. analysis.
According to the M / M / 1 / K model, L~ can be Using K* from (10), the optimal profit NP* can
determined as: be calculated as
p (K + 1)p~:+l
L~ = 1 - p 1 - pr+l (p =/: 1). NP* = ½(X/~ItCTH -- X/C-~OE}2. (ll)
(7) According to (11), NP* will be higher by: increas-
K ing the service rate (/~) associated with production
(p= 1).
at the CCR, increasing the throughput per unit of
sale tCT.), and lowering the carrying cost per unit
Assuming (5) and (6), we can present formula (4) as
(CoE).
follows:
Unfortunately, it is impossible to determine
NP = l~(l - Po) CTH -- L.~CoE. (8) a finite formulas for K* and NP* when p is not
equal to 1. Thus, a numeric analysis is required to
Different sizes of the time buffer (K) affect show how K* and NP* depend on different para-
throughput and operating expenses by changing meters in 13) and (8).
 Z.D. Radorilsky/lnt. J. Production Economics 55 (1998) ! 13- ! 19 117

3. Analyzing the time buffer p value significantly greater than 1; and the number
of customers in the system associated with L~ can-
The goal of this analysis was to identify the not exceed K.
relationships between the optimal time buffer (in The variations of the service rate have a different
terms of the size and net profit) and various charac- impact on the optimal buffer size and the net profit
teristics of the arrival/service process (2,/z, and p), (see Fig. 2). An increase of/~ and simultaneous
as well as cost characteristics (CTH and COE). To do reduction of p cause almost proportional gain of
such analysis, we calculated the values of the opti- NP*. The explanation to this is associated with the
mal buffer size and optimal net profit by varying nature of service rate in this model, which is pro-
one of the characteristics and keeping others as duction rate per unit time at the CCR. The higher
constant. The results of the analysis are described the production rate, the bigger the throughput and
below (we consider p > 1). related to it is net profit.
The variations of the arrival rate affect the opti- The relationship between the optimal buffer size
mal buffer size and the net profit value only if and the cost characteristics ratio (CTH/CoE I ap-
2 approaches It (li = 40), or p is close to 1 (see peared to be almost linear. The optimal net profit
Fig. 1). The closer the arrival rate of inventory to will be higher by increasing throughput and lower-
service (production) rate, the more chances exist ing carrying cost (see Fig. 3). The similar dependa-
that the CCR will experience shortages, and the bility is valid for the optimal size of the time buffer.
more protection (optimal time buffer) is required. Also, we identified an interesting relationship be-
The higher the arrival rate and the farther it from/~, tween K* and the ratio CTH/CoE.Regardless of the
the lower the value of K* (see Fig. 1). The values of CTH and COE values, for the equal ratios the value of
NP* show very low sensitivity to the variations of K* remains the same. This relationship becomes
2. For example, the 100% gain in the arrival rate clear from (10), whe,'e K* is identified for p = i.
2 (from 40 to 80) will increase the optimal net profit Evidently, the same relationship is present when
by 9% only. This low sensitivity is explained by the p>l.
fact that net profit in (8) is affected by t~, Ca-., and The results of the sensitivity analysis can be used
COE, which are constant in this case; Po is almost for planning and scheduling production at the
constant and extremely close to 0 because of the bottleneck, when it is important to develop a

m- - 4760

oi I
m
J 4740

4720

o
4700 =-
o
Z

|.==3 ~- 4680 E
Q,
i o
i,-A--Optimai Buffer Size "= 4660

--II--Optimal Net Profit i

4640

* 4620
0 i I t t ~

40 60 80 100 120 140 160 200

Lambde
Fig. I. Variations of arrival rate lit = 40, C,,,,c,,~ = 61.
118 Z.D. Radovils~./hlt. J. Production Economics 55 (1998) 113-119

25 5000.00

I4500.00
20 [-=A---Oktimal. Buffer Size ~1~"~'~- / 4000.00

g- 15
] 3500.00

3000.00 L
m
2500.00 z

I.g 10 2000.00~
0 O
1500.00

1000.00

500.00

0 ~ ...... ~ .... ~ I ---~ -- I I ! ~. . . . . . . . . . +-- 0.00

4 8 12 16 20 24 28 32 36 39 39.5

Mu

Fig. 2. Variations of service rate (,;. = 40, Cm/Cor = 6).

16 T 4700
I
~ 4600
14 '
~4500

12 i 4400
10 4300 ~.

| 41oo-==
"~ 6 optimal Buffer Size I o
O k--B--Optimal Net Profit ~ 4000

4 t 3900

2 J 3800
0~ ~ .... ~ --t I ~- i ~ ~ i .... 3700
1 1.5 2 3 4 5 6 7 8 9 10

CthlCoe
Fig. 3. Variations of cost characteristics (2 = 50,/, = 40).

production process with the optimal net profit and sophisticated approaches for identifying and ex-
minimum size of the time buffer. ploiting bottlenecks in the production flows. These
steps lead to improvements in operations, through-
put increase, and profit growth. With the approach
4. Summary developed here to identifying the optimal size of the
time buffer, a manager is abie to build the time
The ideas of the TOC, including capacity-con- buffer on the basis of a scientific estimate rather
strained resources and drum-buffer-rope, provide than by trial-and-error. An optimal size of the time
 ZD. Radovilsky/int. J. Production Economics 55 (1998) 113-119 119

buffer specifies a maximum profit received as [5] G. Reimer, Material requirements planning and theory' of
a trade-off between throughput and operating ex- constraints: Can they coexist? A case study, Prod. Inv.
Mgmt. J. 32(4), (1991) 48-52.
penses. If a time buffer of the optimal size requires
[6] W. Renn, M. Stevens, Synchronous manufacturing in
an inventory level which for some reason is imprac- action at GM--Allison transmittion: A change in attitude,
tical, profits close to the optimal can still be pro- 1991 Conf. Proc. American Prod. Inv. Control Society,
duced through a range of buffer size values (closely 1991, pp. 504-511.
allocated to the optimal value). [7] E. Goidratt, J. Cox, The Goal, North River Press,
Croton-on-Hudson, NY, 1986.
These calculations of the optimal size of the time
[8] E. Goldratt, The Haystack Syndrome: Shifting In-
buffer have been done with the assumption of formation Out of the Data Ocean, North River Press,
a single-product system. The next step in such re- Croton-on-Hudson, NY, 1990.
search will be to evaluate the optimal time buffer [9] E. Goldratt, R. Fox, The Race, North River Press,
size for a multi-product system. While this evalu- Croton-on-Hudson, NY, 1986.
ation may require a more complicated mathemat- [~0] M. Umble, M. Srikanth, Synchronous Manufacturing:
Principles for World Class Excellence. South-Western
ical analysis, the general approach will remain the Publishing Co., Cincinnati, OH.
same. [11] M. Umble, Analyzing manufacturing problems using
V-A-T analysis. Prod. Inv. Mgmt. J. 32(2), (1992) 55-60.
[12] R. Chase, N. Aquilano. Production and Operations Man-
agement: Manufacturing and Services, Irwin, Homewood,
References IL, 1995.
[13] M. Spencer, Using "'The Goal" in an MRP, Prod. Inv.
[1] P. Adelman, G. Loucks, Applying theory of constraints in Mgmt. J. 32(4), (1991) 22-27.
a service environment (A case study of performance, [14] R. Askin, C. Standridge, Modeling and Analysis of Manu-
measurement and manufacturing design). 35th Internat. facturing Systems, Wiley, New York, 1993.
Conf. APICS, Montreal, Canada, 1992, pp. 1-10. [15] J. Buzacott, Automatic transfer lines with buffer stacks,
[2] W. Beckett, K. Dang, Synchronous manufacturing: New Internat. J. Prod. Res. 513), 11967) 183-200.
methods, new mind-set, J. Business Strategy i 3(i), 11992) [16] R. Suri, G. Diehl, A variable buffer-size model and its use
53-54. in analyzing closed queuing networks with blocking,
[3] S. Jackson, Teamwork-Redefined by the constraints the- Mgmt. Sci. 32(2), (1986) 206-224.
ory, 1991 Conf. Proc. American Production and Inventory [17] J. Simon, W. Hopp, Availability and average inventory of
Control Society, pp. 193-195. balanced assembly-like flow systems, liE Transitions,
[4] W. Libby, Quaker furniture mal:ages using synchronous 23(2), (1991) 161-168.
manufacturing, APICS - The Performance Advantage. [18] D. Gross, C. Harris, Fundamentals of Queuing Theory,
March 1992, pp. 24-26. Wiley, New York, 1985.