Strategic, Tactical and Operational Decisions

in Multi-national Logistics Networks:

A Review and Discussion of Modeling Issues

Günter Schmidt1

and

Wilbert E. Wilhelm2

1 Information and Technology Management Department

University of Saarland
P.O. Box 151150
D-66041 Saarbrücken, Germany
email: [email protected]

2 Department of Industrial Engineering

Texas A&M University
College Station, Texas 77843-3131
email: [email protected]

May 25, 1999

Revised: August 25, 1999

Accepted for publication on September 8, 1999

by the International Journal of Production Research

Note: this paper should not be disseminated without written permission of the authors.

1
ABSTRACT

The rapidly developing, world-wide marketplace is leading to the geographical dispersion of production,

assembly and distribution operations. This paper deals with three aspects of international logistics networks:

strategic, tactical and operational. The strategic level designs the logistics network, including prescribing

facility locations, production technologies and plant capacities. The tactical level prescribes material flow

management policies, including production levels at all plants, assembly policy, inventory levels, and lot

sizes. The operational level schedules operations to assure in-time delivery of final products to customers.

This paper reviews the literature that deals with strategic, tactical and operational levels and discusses

relevant modeling issues.

ACKNOWLEDGEMENTS

This material is based, in part, upon work supported by the National Science Foundation on Grant DMI-

9500211. The authors appreciate the insightful comments offered by Professor Sila Cetinkaya in our

discussions. The authors are also indebted to three anonymous referees whose comments allowed us to

strengthen an earlier version of this paper.

2
 The rapidly developing, world-wide marketplace has spawned a new allure for the multi-national

company to seek out attractive labor markets to reduce cost, locate close to customers to improve customer

service, and build new markets in developing countries to enhance profits. For example, the recent NAFTA

agreement has led U.S.-based companies to locate assembly plants in Mexico. Companies in Western

Europe have begun to establish similar networks to serve the evolving free markets in Eastern Europe.

However, to be profitable, the multi-national company must deal with a variety of issues. Companies must

consider a number of trade-offs; for example, centralized manufacturing (i.e., production and assembly) to

achieve economies of scale versus decentralized operations that seek to improve customer service by locating

assembly plants closer to customers. However, achieving a favorable trade-off may not be straight-forward

since costs depend upon the countries in which plants are located. For example, the unit cost to

produce/assemble depends upon the labor market in each country. The political environment in a country

determines tariffs and may place unique constraints on operations. Shipments from one country to another

incur border-crossing costs, which may include losses due to monetary exchange rates and to requirements

such as the need to transfer materials to a different carrier. Investment costs may also be related to individual

countries, since each may be willing to offer unique financial inducements to attract new businesses. Risks

must be considered relative to potential profits. A variety of risks face the multi-national company, including

the valuation of currency and the political stability of the countries in which it does business.

Management of multi-national companies involves decisions at strategic, tactical and operational

levels and relates to multi-product, multi-plant logistics systems, which entail production, assembly, and

distribution. The purpose of this paper is to review the literature that deals with strategic, tactical and

operational decisions and to discuss relevant modeling issues. Our goal is to foster insight into issues

relevant to global logistics networks and encourage the development of algorithms that can solve actual,

large-scale problems.

We emphasize that one primary modeling issue is that some aspects of global logistics are difficult –

if not impossible – to represent in a mathematical model (see also Vidal and Goetschalckx (1996)). For

example, political stability and the rate at which a developing market will mature may be difficult to quantify.

3
Thus, the scope of this paper is limited to issues that appear to be quantifiable.

The strategic level prescribes a set of locations where facilities are to be located (i.e., “opened”),

production technologies to be employed at each facility, and the capacity of each plant. Strategic decisions

thus determine the network through which production, assembly and distribution serve the marketplace. Any

model applicable to the strategic level must provide manufacturing capacity to satisfy forecast demand for all

products and observe precedence relationships among assembly tasks. The objective is to maximize total

profit, including the fixed cost of investment to open facilities and the variable cost of manufacturing and

distribution, including “border crossing fees” that might be incurred in transit. We define border crossing

fees to include all costs associated with moving materials from one country to the next, including tariffs and

monetary exchange rates. The strategic level establishes the design of the logistics network and thereby

provides the environment in which tactical and operational levels must perform.

The tactical level prescribes material flow management policies, including production levels at all

plants, assembly policy, inventory levels, and lot sizes. For instance, should finished products be assembled

in large lots and held in centralized warehouses, each of which distributes to a large geographical region, or

should final assembly be performed at numerous locations only on demand? The assembly policy impacts

customer service through the time required to service demand. Thus, it is important to determine a measure

of customer service that can be expected to result at the tactical level and provide this as feedback to the

strategic level in order to improve customer service by providing a more responsive network design.

Some issues such as product design may affect both strategic and tactical decisions. For example,

changing the design of a product to modular form may require a new logistics network to be designed at the

strategic level as well as new material management policies at the tactical level because a modularized

product can be assembled at different locations and customized in order to satisfy unique customer demands.

The operational level schedules operations to assure in-time delivery of final products to customers,

coordinating the logistics network to be responsive to customer demands. In particular, this paper proposes a

scheduling objective of minimizing the throughput time to best serve customers. The associated scheduling

problems must consider the due dates of orders, which must be met in a multi-stage environment. Due dates

4
might be set either internally or externally. Tactical decisions limit the potential locations for carrying out

production and assembly tasks by positioning inventories. Components or subassemblies must be transported

between geographically dispersed sites, incurring cost and requiring time. The question at the operational

level is when to perform a manufacturing task and at which facility so that due dates are met to the fullest

extent possible. The operational level must deal with the environment created by decisions made at the

tactical level, including the availability of components (i.e., parts) and any resulting bottleneck facilities,

which may induce long throughput times, causing due dates to be violated.

The strategic level may deal with a relatively long planning horizon of, say, two to five years since

long lead times are required to construct plants and install processing equipment. A relatively high level of

uncertainty may be associated with demand, political environments, and exchange rates over such a long

planning horizon. The lead time to implement decisions may be shortened by purchasing or leasing existing

facilities or by composing a virtual network by subcontracting other companies. Thus, time frames may not

distinguish decision levels so much as the type of decisions each entails. It can be expected that shorter

planning horizons entail less uncertainty. The tactical level prescribes material flow management policies but

is limited by the network made available by strategic-level decisions. In turn, the operational level is limited

by tactical-level decisions, which, for example, may position inventories in anticipation of forecast demand.

The tactical level deals with a mid-range horizon of, say, 6 - 24 months, forming a bridge between strategic

and operational levels. A model for the operational level should be invoked daily to schedule operations

relative to current information about jobs in process and the status of each.

Global logistics networks differ from domestic systems in a number of qualitative and quantitative

ways. Both must deal with economic factors such as interest rates, market prices, production costs, and

transportation costs, but specific values may be country-dependent and, therefore, more difficult to predict on

a global scale. Establishing the global logistics network is not merely a matter of seeking out the most

favorable labor rates. The global network must be designed and operated to recognize – if not exploit -

import tariffs, export taxes, different income tax rates and duties, duty drawbacks, and transfer prices (e.g.,

Alles and Datar (1998)). The transfer price for a product, the price that the selling “node” in a logistics

5
network charges the purchasing “node,” is a significant determinant of taxes that a company must pay. The

global network must also reckon with political conditions, which also differ by country, and may give rise to

trade barriers and government regulations such as local content rules, which require that specified contributed

value be added within a country. The governments of different countries may offer widely varying

inducements (e.g., tax abatements) to attract new businesses. The location of production and distribution

facilities relative to both suppliers and customers is important to achieving a superior service level and

determines the economies of scale that might be exploited as a competitive strength. Qualitative factors such

as political stability and general infrastructure are difficult to include in mathematical models but may be

crucial to global networks. A number of uncertainties affect the level of risk associated with global logistics

networks. In particular, uncertainties in demand and exchange rates may predominate, although uncertainty

of lead times, supplier reliabilities, and other parameters may also be of concern .

International logistics networks have become familiar in such markets as automobiles, garments and

computers. To develop further insight into relevant issues, we describe two companies that are engaged in

multi-plant, multi-product, multi-country logistics.

The Compaq Computer Company purchases electronic components from suppliers around the world

and assembles circuit cards and personal computers at a number of locations around the world (e.g., Houston,

Texas; Scotland; Brazil; Singapore; and China). Recently, the company decided to change its assembly

policy with the goal of improving customer service. Compaq redesigned its PCs, making modular assembly

of certain features possible. The company adopted a new assembly policy so that, rather than assembling all

PCs at the plant locations and distributing them through a network of warehouses, it will now assemble

components to form a “standard” base that is specialized to customer demand through modular assembly

performed by distributors. The goal of improving customer service has led to a new assembly policy which,

in turn, has led to a modified product design and logistics network.

The MacDonald’s Restaurant chain provides yet another example. The fast-food business is driven

by the need to provide fresh food to customers upon demand. MacDonald’s purchases beef in countries

noted for their beef-raising capabilities, maintains a network of warehousing centers and operates a multitude

6
of restaurants around the world that quickly satisfy customer demands on an assembly-to-order basis.

In some industries (e.g., especially high-technology industries such as computers), rapidly evolving

innovations in technology have shortened product life cycles, placing a premium on quick response and

timely delivery to customers. However, other industries (e.g., process industries) have not experienced such

high rates of product innovation and typically place a premium on capacity utilization. This paper tends to

focus on the high-technology industries, so it should be emphasized that a company’s logistic network must

be designed in relationship to the marketplace in which it competes. Thus, it may be necessary to adapt the

perspective presented in this paper to accommodate the needs of specific industries.

The body of this paper is composed of four sections. Sections 2.0, 3.0 and 4.0 address strategic,

tactical and operational decisions, respectively. Each section reviews related literature and discusses relevant

modeling issues. Collectively, a vast body of literature is related to these three levels, so we focus on the

most relevant work in these reviews. Some papers do not address purely strategic or purely tactical issues as

we define them in this paper, so we have taken some liberty in assigning papers to sections 2.0 and 3.0.

Interest in global logistics networks has grown rapidly in parallel with the global economy. Recent reviews

[Thomas and Griffin (1995); Vidal and Goetschalchx (1997); Ganeshan, Jack, Magazine and Stevens (1999);

and Cohen, and Huchzermeir (1999)], books [Wood, Barone, Murphy and Wardlow (1995); Bramel and

Simchi-Levi. (1997); and Ernst, Kouvelis, and Dornier (1998); Lee (1998); and Tayur, Ganeshan, and

Magazine (1999)], and special issues of Management Science [Graves and Fisher (1997)] and IIE

Transactions [Lee 1997)] describe the state-of-the-art. Section 5.0 gives a brief summary of conclusions and

suggestions for further research.

2.0 THE STRATEGIC LEVEL

This section addresses the strategic level, which designs the logistics network by prescribing facility

locations, production technologies, and plant capacities. The network must integrate with suppliers and

transportation channels to customers. In addition to labor and transportation costs, it must consider other

issues such as the infrastructure, general business environment, closeness to markets and to suppliers, taxes

7
and duties, strategic alliances and joint ventures. In general, the production and assembly of each product

may be accomplished in a set of related facilities, each of which might be located in a different country.

In the first sub-section, we review literature that is relevant to strategic-level decisions. In the second

sub-section, we present a prototype formulation to focus our discussion of relevant modeling issues.

2.1 Literature review. A considerable amount of research has been directed towards strategic aspects of

domestic production/distribution networks [e.g., Cohen and Lee (1985), Cohen, Lee, and Moon (1988);

Cohen and Lee (1988); Graves, Kletter, and Hetzel (1998)]. Historically, research has addressed specific

aspects of the logistics system, for example, distribution [e.g., Geoffrion (1977) and Geoffrion and Powers

(1995)], facility location [e.g., Van Roy and Erlenkotter (1982), Van Roy (1983) (1986), Verter and Dincer

(1995a), and Revelle and Laporte (1996), Francis et al. (1992), Tompkins et al. (1996)] and capacity

expansion [e.g., Luss (1982) and Rajagopalan, Singh, and Morton (1998)]. We focus on the literature that

deals with international networks from a systems perspective.

Verter and Dincer (1992) reviewed the literature related to global manufacturing, concluding that it is

vital to integrate decisions that determine location, capacity acquisition and technology selection, and we

discuss this issue further in a later section. Verter and Dincer (1995a) pointed out that few studies have dealt

with the international aspects of strategic production-distribution models, and Verter and Dincer (1995b)

overviewed the literature that deals with strategic issues relevant to global logistics networks. They

concluded that there exists a limited set of models that addresses these issues and recommended that future

research develop more efficient methods to prescribe optimal configurations.

Cohen and Lee (1989) built upon their work on domestic production/distribution systems, describing

a model for designing an international network and managing material flows within the network. Their

model maximizes after-tax profit subject to constraints on facility capacity, demand requirements, material

balance and government requirements. They proposed a global manufacturing strategy, which combines

strategies for plant operation (regional, consolidation, product focus, process focus and vertical integration),

supply (centralized control, regional control, consolidation and diversification) and distribution

(consolidation, co-location and market service). They did not present a mathematical model, but they

8
apparently did not prescribe plant locations or production technologies. They did, however, describe a

numerical example, demonstrating application of their model to analyze tradeoffs involved in producing

personal computers in an international network.

Kouvelis and Rosenblatt (1997) formulated a mixed integer programming model to prescribe an

optimal design for an international logistics network. Their model focused on policies that individual

countries adopt to attract international trade, including taxation, subsidized financing and local content rules.

They presented numerical examples that demonstrate the sensitivity of the global network design to even

small changes in the policies of just one country.

Taylor (1997) proposed a mixed integer program that prescribes technologies, capacities and tooling

to minimize cost for global manufacturing and assembly. He presented several numerical examples that

demonstrate model application in production sourcing and capital procurement. Taylor’s model focuses on

production technologies but, apparently, does not deal with designing the production/distribution network.

The issues involved in prescribing plant and warehouse locations in a global logistics network may

differ from those traditionally used to locate plants. MacCormack et al. (1994) favored locating in

developing markets or regionalized trade blocks, especially in countries with adequate infrastructure. They

proposed a four-phased analysis: identifying the core competency of the firm; studying the regional

manufacturing configuration, market access, risk management, customer demand, and impact of production

technologies; generating a set of potential sites based on the infrastructure; and ranking potential sites using

quantitative methods. Core competencies form the unique competitive basis of the company and may

involve customer service, product quality and technological leadership, product and/or process attributes,

and/or process flexibility. Rather than using only a traditional cost analysis, Bartmess and Cerny (1993)

recommend that location be based on core competencies of the firm. To identify the best network

configuration, they formulated a return-on-investment model, which considers exchange rates, political

uncertainties, taxes, transfer prices, company costs, and levels of risk, but they described no quantitative

means of evaluating alternative designs.

Global logistics networks involve a number of issues that are not encountered in domestic systems.

9
According to Fagan (1991), global sourcing may improve supplier availability and product quality, but it may

entail the disadvantages of greater risks, higher communication and distribution costs, and larger inventory

buffers that result in longer and more uncertain lead times. Monezka and Trent (1991) described a four-step

procedure for successful global sourcing, and Davis (1992) discussed inconveniences and possible negative

consequences. Ohmae (1989) stressed the importance of strategic alliances in global systems. In addition to

establishing a competitive posture based upon core competencies, Hise (1995) has noted that timing is of

crucial importance in global competition. Evolving trade agreements (e.g., North American Free Trade

Agreement (NAFTA), European Union (EU), Association of South Eastern Nations (ASEAN)) eliminate

trade barriers between countries and tend to homogenize taxes and other rules among countries, so they will

influence the design and operation of future global logistics networks.

Few studies have addressed the uncertainties associated with strategic aspects of global logistics

networks. Kogut (1985) recommended that flexibility, which arises from providing excess capacity in the

form of flexible technologies and using a variety of sources, be used as a response to fluctuations in exchange

rates, government policies, and competition. Lessard and Lightstone (1986) and Carter and Vickery (1989)

discussed risks associated with global logistics and potential means of hedging against them. Hodder and

Dincer (1986) formulated a quadratic program that combines plant location, product flow, and financial

variables. They modeled selling price and fixed costs as random variables but solved their model using an

approximation method. They concluded that modeling exchange rates as random variables leads to programs

that are extremely difficult to optimize. Cohen and Kleidorfer (1993) incorporated location, capacity, product

mix, and material and cash flow in a multi-period stochastic program. They proposed several interacting

stochastic sub-models but gave no formulation or solution details.

Modeling issues. In this section, we present a prototype formulation to focus discussion on modeling issues.

We note that further research is needed to refine our prototype formulation to yield a computationally

effective model that deals with all relevant issues.

We assume that the structure of each product can be represented by an assembly tree in which each node

represents a task (operation) and each arc represents the transit of components between tasks (operations). At

10
denotes the set of tasks that must immediately precede task t. For example, if task t were an assembly

operation, At would represent the set of tasks related to components assembled by task t. Bt denotes the set of

tasks that must immediately succeed task t. In an assembly tree, | Bt | = 1 for all nodes except for the root

node, which represents the finished product and has | Bt | = 0.

We use three types of decision variables to prescribe the strategic logistics network. Selecting from a

set of K possible locations, binary decision variable Zk =1 if location k (k ∈ K) is “opened” and 0 otherwise.

To prescribe the technology for each operation, binary decision variable Ykmrt = 1 if resource r is employed at

location k to complete task t on product m and 0 otherwise. Finally, decision variable Xkk’mrt prescribes the

number of products of type m for which task t is performed using resource r at location k and then transported

to location k’ for the following operation. To represent the sale of finished products to customers, we

specialize the indices of variable Xkk’mrt to X wcmrt 0 in order to denote the number of finished products of type

m transported (i.e., task t specializes to “transport” t0) from warehouse w (i.e., location k specializes to the

location of warehouse w) to customer c (i.e., location k’ specializes to customer location c) using

transportation mode r (i.e., resource r specializes to transport mode r).

Before presenting the prototype formulation, we summarize notation for reader convenience.

Notation

Indices
c customers c ∈C
k locations k∈ K
m product types m ∈M
r resource types r ∈R
t tasks required to produce product m t ∈ Ψm
w warehouses w ∈W

Parameters
akmrt amount of resource r required to perform task t on product p at location k
C$kI investment cost to open location k
C$ B
kk ' mt border crossing cost from location k to k’for 1 unit of product m after task t is performed
C$ kmrt
P
cost to process task t on product m using resource r at location k
C$ R
kmrt fixed cost to provide resource r at location k to complete task t on product m
C$kkT ' mt cost to transport one unit of product m from location k to k’after task t is performed
Dcm number of products of type m demanded by customer c

11
Qm upper bound for production of product m = ∑c
Dcm

Ukmrt availability of resource r at location k to perform task t on product m

Πwcm the price that customer c must pay to purchase one unit of product m supplied by warehouse w

Sets
At immediate predecessors of task t
Bt immediate successors of tasks t (|Bt| = 1 for an assembly tree structure)
C customers
K locations
Kt locations at which task t could be performed if opened Kt ⊆ K
M product types
R resource types
Rt resource types that can complete task t
W warehouses
Ψm tasks required to produce product m

Decision Variables
Xkk’mrt number of products of type m for which task t is performed using resource r at location k and
then transported to location k’ (t’∈ Bt )
Ykmrt 1 if resource r is employed at location k to complete task t on product m, 0 otherwise
Zk 1 if location kt is “opened”, 0 otherwise

note:
X wcmrt 0 ⇒ Xkk’mrt
number of finished products of type m transported (t0 ⇒ “transported”)
from warehouse w (w ⇒ location k) to customer c (c ⇒ location k’) using mode r (i.e., resource r)

We now present our prototype formulation of strategic-level decisions.

Objective: Maximize Profit

Max Z = ∑
wcmr
Πwcm X wcmrt 0

- ∑ C$kI Zk - ∑ C$ kmrt
R
Ykmrt - ∑ ( C$kkB ' mt + C$ kmrt
P
+ C$kkT ' mt ) Xkk’mrt (1)
k kmrt kk ' mrt

Subject to:

∑ wr
X wcmrt 0 = Dcm c ∈ C, m ∈ M, t0 ⇒ transport (2)

∑ ∑
k "∈ K t " r ∈ Rt "
Xk”kmrt” - ∑ ∑ ∑
t '∈ Bt k '∈ K t ' r ∈ Rt
Xkk’mrt = 0 m ∈ M, t ∈ Ψ m, k ∈ Kt, t” ∈ At (3)

Xkk’mrt ≤ Qm Z k m ∈ M, t ∈ Ψ m, k ∈ Kt, t’∈ Bt, k’∈ Kt’, r ∈ Rt (4)

∑
k '∈ K Bt
akmrt Xkk’mrt ≤Ukmrt Ykmrt m ∈ M, t ∈ Ψ m, k ∈ Kt , r ∈ Rt (5)

12
 ∑
k ∈K t
Zk ≥1 m ∈ M, t ∈ Ψ m (6)

Ykmrt ≤Zk m ∈ M, t ∈ Ψ m, k ∈ Kt, r ∈ Rt (7)

Ykmrt ∈ {0, 1} m ∈ M, t ∈ Ψ m, k ∈ Kt , r ∈ Rt (8)

Zk ∈ {0, 1} k ∈K (9)

Xkk’mrt ≥ 0, Integer m ∈ M, t ∈ Ψ m∪ { t0 }, k ∈ Kt , k’∈ Kt’, r ∈ Rt (10)

The objective (1) is to maximize profit. A revenue of Πwcm accrues for each of the X wcmrt 0 sales. To

calculate profit, we subtract the fixed cost to open location k (denoted by C$kI Zk); the fixed cost of providing

resource r at location k to complete task t on product m (denoted by C$ kmrt

R
Ykmrt); and the variable costs

associated with border crossing, manufacturing operations and transportation (denoted by ( C$kkB ' mt + C$ kmrt
P
+

C$kkT ' mt ) Xkk’mrt).

Inequality (2) assures that the demand of each customer for product m, Dcm, is satisfied. Equality (3)

invokes a flow balance, assuring that the flow of product m into task t (at location k) from each immediately

preceding task t” ∈ At (at location k”) equals the flow of product m from task t to immediately succeeding

task t’∈ Bt (at location k’). Inequality (4) relates X and Z decision variables, assuring that task t on product

m cannot be performed at location k if that location is not opened. Inequality (5) relates X and Y decision

variables, invoking the capacity of the resource r, which is prescribed to perform task t on product m at

location k. The capacity of resource r provided at location k for task t on product m may also be prescribed,

for example, by defining r to denote resource type as well as its capacity (availability). Setting a particular

Ykmrt = 1 would then select resource r and its availability Ukmrt. Of course, this convention would require

additional constraints of the form ∑ ∑ Ykmrt = 1 in which r is summed over the set of resource types that can

complete task t, Rt, as well as the set of capacities for each resource type. The model would then prescribe

one resource to perform task t on product m at location k as well as the capacity of that resource. Inequality

(6) requires that at least one facility, which can perform task t for product m, be opened. Inequality (7) relates

Y and Z decision variables, assuring that location k employs technology (i.e., resource) r only if it is opened.

13
Finally, constraints (8) – (10) impose binary and integer requirements.

Formulation (1)-(10) exhibits an interesting structure. Equalities (2) and (3) invoke task precedence

relationships and impose restrictions related to network flow constraints. Equality (3) requires that the flow

of product m from task t to its immediate successor, t’, equal the flow from immediate predecessor t” to t.

However, the assembly tree structure requires a constraint of type (3) for each t” ∈ At, so (3) does not

represent a classical network flow problem. It does, however, represent bill-of-material (BOM) relationships

associated with assembly operations. With Zk variables fixed to binary values, inequality (4) becomes an

upper bound for flow on an arc in this network. Constraints can be decomposed to form an independent flow

problem involving constraints (2), (3) and (4) for each product. Fixing Ykmrt variables to binary values

introduces resource-capacity limitations on the performance of task t on product m at location k. Alternative

production configurations could be represented by appropriate constraints. For example, if resource type r

could be used to perform all tasks assigned to location k, then (5) could be formulated as

∑ ∑ ∑
m∈ M t
akmrt Xkk’mrt ≤Ukr Ykr k ∈ K, r ∈ R (5’)
k '∈ K Bt

in which the summation on t includes the subset of tasks in set Ψ m that can be performed at location k by

resource type r.

Since international logistics networks frequently involve production as well as assembly, the

structure of the strategic-level model does not conform to traditional production models. Constraints (1)-(3)

evidence this complication in that they do not form pure network flow constraints.

Because the strategic level may involve a relatively lengthy planning horizon, decision makers would

typically not have enough information to specify all parameters with certainty. Strategic-level decisions

entail a relatively high level of uncertainty associated, for example, with demand, political environment, and

exchange rates. In fact, it may be difficult even to specify the set of products, M, in great detail over a long

planning horizon since most products tend to have short life cycles in today’s marketplace. Thus, it may be

necessary to treat parameters like Dcm , akmrt , Ukmrt , Πwcm , C$kkB ' mt , C$ kmrt
P
, and C$kkT ' mt as random variables.

These considerations lead to a stochastic, integer formulation.

14
 Another issue involves time dependency. The prototype formulation (1)-(10) is a static model in that

the decisions, which are prescribed at the time the model is implemented, fix the design of the logistics

network for the entire planning horizon. In practice, it may be possible to add facility locations and plant

capacity incrementally over time. In such a case, the parameters and decision variables in formulation (1)-

(10) would have to be amended by adding a subscript to designate time period, for example, years 1, 2, … , 5.

This consideration would lead to a stochastic, time-dependent, integer program, a type of model that presents

exceptional computational challenges.

Depending upon the application, other issues may also be relevant. For example, the transportation

system employed may have a cost structure that is most appropriately modeled as a concave function of the

quantity transported. Furthermore, it may be of interest for the model to prescribe whether flexible or

dedicated technologies should be employed [e.g., Cetinkaya and Falkner (1995) and Benjaafar and Gupta

(1998)]. In addition, it may be important to model transfer pricing explicitly, so that the network design

reflects this important determinant of taxes. Typically, transfer pricing introduces nonlinearities and results in

a nonconvex model [Vidal and Goetschalckx (1998)]. Transfer pricing may be an issue in strategic, tactical

and operational levels.

Finally, we note that this prototype formulation is related to Geoffrion’s (1974) model for designing

multi-commodity distribution systems. He successfully solved his model through a specialized application of

Benders decomposition. Geoffrion’s model deals with a set of plants that produce finished products, which

are then stored in distribution centers for shipment to various customer zones. His model focused on

distribution, prescribing locations for distribution centers and the capacity of each. The prototype

formulation (1)-(10) extends Geoffrion’s focus to allow parts and subassemblies to be processed in a series of

plants, forming a larger, multi-echelon logistics network. Geoffrion’s model does not reflect suppliers,

assembly BOMs, stochastic aspects, time dependencies or other considerations important to international

logistics networks. We now turn attention to the tactical level.

3.0 THE TACTICAL LEVEL

15
The tactical level prescribes material flow management policy, including production levels at all plants,

assembly policy, inventory levels, and lot sizes. It deals with material flow from suppliers to production

facilities, to assembly plants, through warehouses and on to customers. The strategic level must deal with a

number of uncertainties, and the tactical level, which deals with the time horizon of, perhaps, 6-24 months,

affords an opportunity to refine production plans to compensate for political and financial processes as they

unfold. Products ordered for delivery in the near term may be known with certainty but, still, later periods in

the tactical horizon may be subject to some degree of uncertainty.

Customers demand highly individualized products within short lead times, so product variety is a

primary issue at the tactical level. Customer service is a function of the assembly policy, which can, for

instance, require finished-product inventories or assemble-to-order operation. If immediate delivery of

finished products is critical, customized products must be assembled and held in inventory [e.g., Schmidt and

Nahmias (1985) and Rosling (1989)]. In addition, production lot size must be determined at the tactical level

[e.g., Afentakis, Gavish, and Karmarkar (1984); Afentakis and Gavish (1986); Atkins Queyranne, and Sun

(1992); Axsater and Nuttle (1987)]. In assemble-to-order systems [e.g., Glasserman and Wang (1996),

Schraner (1996), and Zhang (1996)], in which only components - not products - are inventoried, the challenge

is to respond to consumers within pre-specified times with no finished-product inventory. To achieve this,

product designs can be based on interchangeable modules that are assembled into different final products on

demand, customizing them as required. To be profitable, a company must achieve a favorable trade-off

between service level (i.e., leadtime) and inventory cost (i.e., fill rates).

The tactical level considers both production and transportation. Such decisions are often made

separately, but that is possible only when the decisions have few interdependencies. The question is how to

assign production to plants in order to minimize production, changeover, holding, and transportation costs. In

the multi-plant environment, each assembly operation must be assigned to one plant from among the

candidates that offer the required technological capability and production capacity. Even though the strategic

level develops a long-range plan for assignment of operations to locations, this decision must be refined at the

tactical level as demands, exchange rates, political environment, and other factors become better defined.

16
 The first sub-section in this section reviews literature relevant to tactical-level decisions. The second

sub-section discusses relevant modeling issues.

3.1 Literature review. Cohen, Fisher and Jaikumar (1989) presented a preliminary model for coordinating

procurement, manufacturing, and distribution in global networks. They discussed many of the factors that are

germane to international networks, including economies of scale, duties and tariffs, currency exchange rates,

global sourcing, and country-specific considerations such as local sourcing rules and tax rates. Their model

incorporates many of these factors, maximizing after-tax profits subject to constraints on material supply,

material requirements, plant shipments, market demand, market cash flows, plant capacities and local content

rules. They also discussed techniques for dealing with random currency fluctuations. Marketing and

financial considerations are the focus of this model, which is a mixed integer program with nonlinearities

introduced by variables that determine transfer prices and allocate overhead to plants. Their model is based

on a three-level network, composed of vendors, plants and markets, and does not deal with assembly, plant-

to-plant transfers or warehousing.

Lee and Billington (1992) discuss potential problems and opportunities associated with managing

supply chain inventory. Arntzen, Brown, Harrison and Trafton (1995) described practical issues encountered

by Digital Equipment Corporation in its international logistics network and discussed application of a large-

scale, mixed integer linear program to minimize the sum of fixed and variables costs. The latter included the

costs of production, inventory, shipping, and duties.

Vidal and Goetschalckx (1996) modeled the impact of exchange rates and supplier reliability on

system configuration in a mixed integer program (MIP) and demonstrated the sensitivity of the global

logistics network design to these parameters. They classified relevant factors as those that can be modeled

with a high degree of accuracy, those that can be modeled adequately by invoking assumptions, and those

that are very difficult to model. They emphasized the need to address uncertainties and noted that there is

currently no generally accepted quantitative methodology for dealing with global logistics networks. Vidal

and Goetschalckx (1998) described a detailed model to maximize after-tax profits, considering transfer prices

and transportation costs. They presented a heuristic to solve their model, which is not convex.

17
 Canel and Kumawala. (1997) proposed a MIP model to optimize the global supply chain for a single

product, however it does not prescribe transfer prices and transportation costs. Dasu and de la Torre (1997)

modeled a multi-national company with partially owned subsidiaries that must compete with each other as

free trade develops, determining production quantities for each plant and sales prices that are responsive to

exchange-rate fluctuations. They presented a solution methodology and a numerical example, demonstrating

application to the fiber industry in Latin America.

Few studies have addressed the uncertainties associated with tactical aspects of global logistics

networks. Cohen and Lee (1988) formulated stochastic sub-models for a serial production process but

concluded that optimizing all sub-models is not tractable, so they proposed a heuristic solution method.

Cohen and Lee (1988) claimed that many uncertainties and qualitative factors may be analyzed through the

specification of different scenarios and sensitivity analysis. They noted that there is currently no formal and

consistent way to represent BOM constraints (i.e., assembly operations as in constraints (2)-(4)). Cohen,

Fisher, and Jaikumar (1989) fixed transfer prices and allocated overhead to reduce their model to a tractable

MIP. Kogut and Kulatilaka (1994) proposed a model to prescribe the shifting of production between plants

in two countries. They developed a stochastic dynamic program to deal with randomly fluctuating exchange

rates. Cole (1995) proposed a more limited MIP model of production and distribution that considered

stochastic safety stock under normally distributed demand.

3.2 Modeling issues. Considering formulation (1)-(10) as a prototype for a tactical-level model, we see that

the strategic level fixes Zk and Ykmrt decision variables relative to the description of products, demands and

exchange rates available when the model is implemented. Decision variables Xkk’mrt represent the material

flow management decisions emphasized at the tactical level. More information is available to the tactical

level as random processes unfold, so the relative levels of uncertainty are less than in the strategic level. The

tactical level makes flexible use of the plant capacities made available by strategic-level decisions.

Basic issues that must be modeled at the tactical level include:

(1)what is the optimal assignment of operations to plants, taking into account plant capacities that

result from strategic decisions and actual product demand that develops,

18
 (2)what is the optimal assembly policy for each product,

(3)what is the optimal inventory level for each subassembly, taking into account the coordination of

shared inventories and interdependent demand, and

(4)what is the optimal lot size for assemble-to-order components.

Relative to formulation (1)-(10), parameters and decision variables must be amended by adding a subscript to

designate time period, for example, months 1, 2, … , 18. This consideration again leads to a challenging,

stochastic, time-dependent, integer program.

Many of the basic issues seem to have been well structured by prior work and research needs are

widely recognized. In a critical review of global supply chain models, Vidal and Goetschalckx (1997)

emphasized the need for more comprehensive models as well as specialized solution methods. They noted

that there “exists a lack of research on MIP models for the strategic design of global supply chain systems”

and that “most models do not include single sourcing, inventory costs, and BOM constraints,” which are

fundamental to assembly systems. They also pointed out that “Most research addresses a single component

of the overall production-distribution system, such as purchasing, production and scheduling, inventory,

warehousing or transportation. To date there exists little research that addresses the integration of such single

components into the overall supply chain. “… there exist many opportunities for developing more

comprehensive global supply chain models that include BOM constraints, more stochastic factors and

qualitative factors that are important in the global environment.” They also observed that solutions methods

have been limited to Benders decomposition [Geoffrion (1974)] and factorization methods [Brown and Olson

(1994)], and identified a primary research need as the development of specialized methods of solution.

4.0 THE OPERATIONAL LEVEL

The operational level schedules operations to assure in-time delivery of final products to customers. In

contrast to the strategic and tactical levels, little research has been directed towards scheduling multi-echelon,

multi-national operations. One reason for this may be that scheduling problems may be viewed as specific to

the degree of management’s centralization (or decentralization), the performance measures used to evaluate

managers, and the process configuration (e.g., single machine, parallel machines, flow shop, job shop,

19
cellular organization). Nevertheless, the operational level can significantly affect customer service, so it is an

important component in providing a unified logistics system.

Our discussion focuses on coordinating schedules for multi-echelon assembly systems in which each

process may be located in a different country, and we describe application of optimization to coordinate

material flow in such a system. Our approach to coordination is represented by constraints; for example, the

strategic level determines system capacity and the tactical level prescribes material flow, setting the stage for

scheduling by positioning inventories. We propose use of optimization as a means of fulfilling a particular

customer-service objective to the extent possible, given the constraints imposed by strategic and tactical

levels. To focus on multi-echelon coordination, we assume implicitly that each process is performed by a

unique workstation dedicated to a specified set of products. Our presentation may be adapted to address

specific industries that might have different needs.

Companies having dispersed locations in several countries may gain many advantages by reducing the

time and cost associated with production and delivery. On the other hand, coordinating the underlying

logistics network becomes more difficult. Demand for products occurs at points in time when customers

place orders. The focus at the operational level is on where and when to assemble components to minimize

the time interval from order arrival to order delivery at the customer's site. We intend for our terminology to

be generic so that an “operation” can represent the processing of a part, a component, a subassembly, a

module, or the end product in batches of size one or larger.

It is more difficult to answer these questions for multi-component products that are assembled in

networks than it is for single-component products that require no part coordination. In environments that

involve multiple components, both serial and parallel operations must be considered. More generally,

connections between processing facilities in networks must be allowed arbitrary structure.

The first sub-section in this section reviews literature relevant to operational-level decisions. The

second sub-section discusses modeling issues.

20
4.1 Literature review. We know of no research that has focused specifically on scheduling in multi-national

logistics networks. However, the literature describes prior research on two topics related to the operational

level: scheduling assembly operations and scheduling in networks. We will give a short review of each topic.

Little (1994) reviews case studies for scheduling assembly systems. Russell and Taylor (1985)

studied scheduling in an assembly shop, evaluating dispatching policies based on priority rules. Each job

consists of several manufacturing and assembly operations, which define a tree precedence structure. The

shop configuration consisted of a number of machine centers, which were dedicated to manufacturing

operations, and one machine center, which was assigned only assembly operations. Russell and Taylor

treated labor as a scarce resource that had to be shared between the machine centers. They considered no

transportation times nor alternative machine centers for operations. They analyzed several objectives related

to job completion time (e.g., mean flow time, mean tardiness and percent jobs tardy) and studied the impact

of priority rules in meeting these different objectives.

Phillips, Stein and Wein (1997) studied scheduling in networks. They assumed that each job consists

of a single operation, which can be performed by any machine in a set of identical (unrelated) machines. The

machines are connected by a network defining distances between machines by shortest paths between pairs of

nodes in the network. Each job has a certain starting node from which it proceeds. They investigated two

objectives: makespan and sum of completion times. To solve the problem, they devised approximation

algorithms.

Flipo and Finke (1997) investigated a similar problem in network scheduling. They also assumed

that a job consists of a single operation and dealt with networks in which nodes relate to production facilities,

warehouses and customers, and arcs represent transportation links. The problem combined production and

distribution in a setting in which each facility had several production lines dedicated to certain product types

with different production rates. Moreover, they considered changeover times if products of different types

were produced in sequence on the same line. They modeled the system as a network flow problem with the

objective of minimizing the sum of production, changeover, holding, and transportation costs.

Huang and Xu (1998) investigated a multi-stage, multi-item network scheduling problem. They

21
considered storage space limitations and the production capacities of workstations. Their model minimized

the total cost of holding, processing, transporting, and schedule delay by prescribing the times at which jobs

enter the system and the time-dependent processing route for each job. They adapted a network algorithm to

solve the problem, invoking the simplifying assumption that operations need not be coordinated.

4.2 Modeling issues. Tactical-level decisions position inventories for use by the operational level. The parts

and subassemblies needed for each assembly operation must be available before the operation can begin.

This interface between the tactical and operational levels of assembly systems has been studied by Chen and

Wilhelm (1993) (1994) (1997).

A model of the operational level must be based on structural assumptions concerning the products

and the logistics network. For example, one may assume, without loss of generality, that each location has

only one workstation. If a location actually has more than one workstation, each may be treated as a separate

location. We denote the graphical representations of the logistics network and the products by G1 and G2,

respectively. Nodes of G1 represent locations (workstations); and edges, transportation links between them.

Edge weights give information about the distances between the respective locations. The questions of when

and where to move the products in the network are critical for achieving efficient resource utilization.

Every product has an associated graph G2, which describes its BOM and feasible sequences of

assembly. G2 has a special structure, which is an in-tree. An in-tree consists of three kinds of nodes: nodes

without predecessors (leaves), nodes without successors (root), and nodes with predecessors and successors

(intermediates). Intermediate nodes have an arbitrary number of predecessors but only a single successor; a

leaf has no predecessor and a single successor node, and a root has an arbitrary number of predecessor nodes

but no successor node.

Each node in an in-tree can be assigned to a layer. Nodes belonging to the same layer have the same

number of components in the backpath to the root node. An in-tree has as many layers as there are

components on the longest path from a leave node to the root. The layers describe the sequence in which

components must be made available to assemble the product. We use the term “component tree” to indicate

an in-tree with a given layer structure.

22
 Each product j is made from a set of components Sj = {Silj | i = 1, ..., n(l), l = 1, ..., n(j)} that must be

joined by assembly operations. n(l) denotes the number of components in layer l and n(j) is the number of

layers that compose product j. Component Silj is assembled on layer l by joining its predecessor components

Si,l-1,j from layer l-1. We denote the corresponding assembly operation by Tilj. Component S1,n(j),j represents

the finished product j as well as the root of the component tree.

Figure 1 gives an example of graphs G1 and G2. The logistics network depicted consists of nine

locations (k) that are connected by ten transportation links (k, k'); edge weights are not shown. Product j

consists of five components Silj related in three layers of the component tree and can be assembled if the

constraints represented by the arcs are observed. Node S13j represents the final product.

Insert Figure 1 here

Locations in the multi-national network may be in different time zones so that they might not all be

available simultaneously in certain time intervals. Consider, for example, locations in Germany and

California that operate for two eight-hour shifts, starting at 6am local time. Normalizing the time scale to

Pacific time, the California location is active from 6am to 10pm and closed from 10pm to 6am. The German

location is active from 9pm to 1pm and down between 1pm and 9pm Pacific time. If a location operates

three shifts so that it is continuously active, downtime intervals are not defined.

Given a customer order, the operational-level problem is to determine where and when operations

should be performed to meet given constraints and optimize the objective function. Two problem types exist:

(1)The order sequencing problem (OSP): locations (workstations) for assembly operations are fixed

and there is a unique location (workstation) for each operation. This leads to an OSP in which one

must determine the time when each operation is performed at each location in which the operation is

performed.

(2)The network scheduling problem (NSP): each assembly operation can be performed at several

candidate locations (workstations). This leads to a NSP in which one must decide where an

operation is to be performed as well as when.

23
 We use the term “job” to indicate a product that is ordered by some customer. Once an order is

placed, it must be completed according to the product's component tree. A collection of jobs represents

customer orders (i.e., demand) that must be fulfilled during the planning period. We assume that the

following data are known at the operational level:

(1) component trees for all jobs (i.e., products),

(2) processing time pilj for assembly task Tilj

(3) transport time qi,l(l+1),jkk' between locations k and k',

(4) order arrival date rj for job j,

(5) due date dj for delivery of job j,

(6) intervals of location downtimes: the v-th interval associated with location k is given by [Bvk, Fvk].

We use Cj to denote the completion time of job j, that is, the time at which job j is delivered to the customer.

Different objectives related to completion times may be used at the operational level. It might be necessary to

deliver the collection of jobs as quickly as possible. In this case, the scheduling objective is to Minimize

Cmax = Minimum maxj{Cj}. On the other hand, if it is important to avoid late delivery, the scheduling

objective should be to minimize a function of the total tardiness of all jobs, where tardiness for job j is

defined as Dj = max{Cj - dj, 0}. Other scheduling objectives relate to the flow (throughput) time Fj = Cj - rj.

For example, it is often necessary to minimize the sum of the flow times.

Schedules must also observe different constraints. The most basic ones are:

(C1) Each operation can be assigned to only one location at a time.

(C2) Each location can only process one operation at a time.

(C3) Jobs can only be processed in active intervals.

The OSP and NSP are to find a demand-satisfying schedule that minimizes a non-decreasing real

function of job completion times. Standard mathematical programming formulations can be used to give a

formal description of these problems [Blazewicz, Dror, and Weglarz (1991)].

4.2.1 The order-sequencing problem (OSP). In the OSP, the location (workstation) k at which each

assembly operation must be performed is pre-determined. We use Tiljk to denote that the assembly operation

24
associated with component Silj must be performed at location k. All data describing the logistics network can

now be represented within the component tree. We use the term “assembly tree” to indicate a component tree

that incorporates information about assembly locations. Figure 2 depicts an assembly tree for an order related

to the component tree given in Figure 1b.

Insert Figure 2 here

The example shows that five operations must be completed over three layers at five different

locations. Locations 1, 2 and 3 complete operations T11j1, T21j2, and T22j3, respectively; location 4 assembles

components S11j and S21j. Location 5 completes the final assembly operation T13j5, combining subassemblies

S12j and S22j.

In order to represent the input data for the OSP, we assign weights to the nodes and arcs in the

assembly tree. A node weight represents an operation processing time and an arc weight represents the

transport time between two locations. Operation Tiljk requires processing time piljk. After operation Tiljk is

completed at location k on layer l, the assembled component must be transported to location k' on layer l + 1

for the next operation, requiring transport time qi,l(l+1)j,kk'. Figure 3 extends the example shown in Figure 2,

adding processing and transport times.

Insert Figure 3 here

The notation used to describe the OSP may be simplified because there is only one candidate location for

each operation. Each operation is already defined uniquely by index ijl. Let Aj be the set of ordered pairs of

operations (Tilj,Ti,l+1,j) for each job j. Let Ek be the set of pairs of operations to be processed at the same

location k. Due to technological and capacity constraints, pairs from Aj and Ek must not overlap. Operation

Tilj requires processing time pilj. The primary decision variable, xilj, prescribes the starting time for each

operation in an optimal schedule.

The objective of an OSP may minimize some function of job completion times. Constraints must

ensure that operations observe precedence relationships, that each workstations performs one operation at a

time and that a job can only be processed by the assigned location during an active interval. Finally,

constraints must assure that each job is started but not before its release time rj.

25
4.2.2 The network-scheduling problem (NSP). In the NSP, each assembly operation can be performed at a

number of candidate locations (workstations). In this case, a job is not represented uniquely by an assembly

tree. Component trees and the logistics network define the input data for the NSP. We introduce decision

variable xiljk to denote the start time of the i-th operation of job j on layer l at candidate location k.

The objective of an NSP may minimize some function of job completion times. Constraints must

assure that operations observe precedence relationships, that each location processes no more than one

operation at a time and that jobs are processed only during active intervals. Such a formulation is general

enough to subsume most classical, multi-stage production scheduling problems, which invoke assumptions

like the following ones. A job shop means that there may be a different sequence of operations for each job

and a flow shop means that the sequence is the same for all jobs.

(a) One operation, one processor for each layer:

Sequential job shop: l = n(j), i = 1,

Sequential flow shop: l = n(j), i = 1,

Single machine: l = 1, i = 1.

(b) One operation, multi processors for each layer:

Flexible job shop: l = n(j), i = 1,

Flexible flow shop: l = n(j), i = 1,

Parallel machines: l = 1, i = 1.

(c) Multi operations, one processor for each layer:

General job shop: l = n(j), i = n(l),

General flow shop: l = n(j), i = n(l),

Single machine: l = 1, i > 1.

(d) Multi operations, multi processors for each layer:

Gen-Flex job shop: l = n(j), i = n(l),

Gen-Flex flow shop: l = n(j), i = n(l),

Parallel machines: l = 1, i > 1.

26
Both OSP and NSP models can address further details, including, for example, the following features:

(a) If more than one component type is processed at a location, change-over time sjj'k could be added.

(b) A location may hold some inventory of assembled components that could be used to satisfy a demand,

with essentially zero processing time.

(c) Components must be batched for transport between locations.

5.0 CONCLUSIONS

This paper reviews the literature that deals with strategic, tactical and operational decisions related to multi-

national logistics networks. The strategic level designs the logistics network, including prescribing facility

locations, production technologies and plant capacities. The tactical level prescribes material flow

management policies, including production levels at all plants, assembly policy, inventory levels, and lot

sizes. The operational level schedules operations to assure in-time delivery of final products to customers.

These levels interact in several ways. First, higher levels establish constraints that affect performance at

lower levels. For example, the strategic level determines process capacity, and the tactical level positions

inventories for use by the operational level. Second, each level addresses a particular time frame, and these

time frames must be integrated to assure seamless customer service. For instance, the operational level must

assure that short-term material flow meets customer-service goals, coordinating strategic and tactical levels

appropriately. Thus, an approach that unifies these three levels is necessary to design and operate a

competitive, global logistics network.

This paper also discusses modeling issues that are relevant to each decision level. It is intended that

these issues be generic, but different industries may place varying degrees of emphasis on specific issues.

Thus, different industries may need to adapt the models described in this paper to allow them to compete

most effectively in particular marketplaces. Finally, we observe that some aspects do not lend themselves to

quantitative modeling and must be addressed by the experience and judgement of capable managers.

It is our hope that this work will foster insight into issues relevant to multi-national logistics

problems and encourage development of algorithms that can solve actual, large-scale problems. Our ongoing

27
research is directed towards formulating specific models of the type described in this paper and towards

developing effective solution methods.

6.0 REFERENCES

Afentakis, P. and Gavish, B. (1986) Optimal Lot-Sizing Algorithms for Complex Product Structures.
Operations Research, 34 (2), pp. 237-249.

Afentakis, P., Gavish, B., and Karmarkar, U. (1984) Computationally Efficient Optimal Solutions to the Lot-
Sizing Problem in Multistage Assembly Systems. Management Science, 30 (2), pp. 222-239.

Alles, M. and Datar, S. (1998) Strategic Transfer Pricing, Management Science, 44 (4), 451-461.

Arntzen, B.C., Brown, G.G., Harrison, T.P., and Trafton, L.L. (1995) Global Supply Chain Management at
Digital Equipment Corporation, Institute for Operations Research and the Management Sciences, 25 (1), 69-
93.

Atkins, D., Queyranne, M., and Sun, D. (1992) Lot Sizing Policies for Finite Production Rate Assembly
Systems. Operations Research, 40 (1), pp. 126-141.

Axsater, S. and Nuttle, H.L. (1987) Combining items for Lot Sizing in Multi-Level Assembly Systems.
International Journal of Production Research, 25 (6), pp. 795-807.

Bartmess, A., and K. Cerny (1993) Building Competitive Advantage through a Global Network of
Capabilities, California Management Review, Winter, pp. 78-103.

Benjaafar, S. and Gupta, D. (1998) Scope Versus Focus: Issues of Flexibility, Capacity, and Number of
Production Facilities, IIE Transactions, 30, 413-425.

Blazewicz,J., Dror,M., Weglarz,J. (1991) Mathematical programming formulations for machine scheduling:
A survey, European Journal of Operational Research, 51, pp. 283-300.

Blazewicz,J., Ecker,K., Pesch,E., Schmidt,G., Weglarz,J. (1996) Scheduling Computer and Manufacturing
Processes, Springer-Verlag, New York.

Bramel, J., and D. Simchi-Levi. (1997) The Logic of Logistics, Springer Series in Operations Research,
Springer-Verlag, New York.

Brown, G.G., and M.P. Olson. (1994) Dynamic Factorization in Large-Scale Optimization, Mathematical
Programming, 64, pp. 17-51.

Canel, C., and B.M. Kumawala. (1997) Multi-Period International Facilites Location: An Algorithm and
Application, International Journal of Production Research, 35 (7), pp. 1891-1910.

Carter, J. R., and S. K. Vickery (1989) Managing Volatile Exchange Rates in International Purchasing,
Journal of Purchasing and Materials Management, Winter, pp. 13-20.

Cetinkaya, S., and C. H. Falkner (1995) Multi-Machine Replacement Decisions in the Context of Flexible
Manufacturing, International Journal of Manufacturing System Design, 2 (4), pp. 277-29

28
Chen, J.F., and Wilhelm, W.E., "An Approach for Optimizing the Allocation of Components to Kits in
Multi-Echelon Assembly Systems," Naval Logistics Research, 41 (2), 1994, 229-256.

Chen, J.F., and W.E. Wilhelm, "An Evaluation of Heuristics for Allocating Components to Kits in Small-
Lot, Multi-Echelon Assembly Systems," International Journal of Production Research, 31 (12), 1993, 2835-
2856.

Chen, J.F., and W.E. Wilhelm, "Kitting in Multi-Echelon, Multi-Product Assembly Systems with Parts
Substitutable," International Journal of Production Research, 35 (10), 1997, 2871-2897.

Cohen, M.A., Fisher, M., and R. Jaikumar. (1989) International Manufacturing and Distribution Networks:
A Normative Model Framework. Managing International Manufacturing, K. Ferdow (ed), Elsevier Science
Publishers, North Holland, pp. 67-93.

Cohen, M. A., and A. Huchzermeir, “Global Supply Chain Management: A Survey of Research and
Applications,” Chapter 21, Quantitative Models for Supply Chain Management, Tayur, S., Ganeshan, R., and
M. Magazine (eds), Kluwer International Series, Kluwer Academic Publishers, Boston, 1999, 669-702.

Cohen, M.A., and P.R. Kleindorfer (1993) Creating Value through Operations: The Legacy of Elwood S.
Buffa, Perspectives in Operations Management(Essays in Honor of Elwood S. Buffa), R. K. Sarin (ed),
Kluwer Academic Publishers, Boston, MA, pp. 3-21.

Cohen, M. A., and Lee H. L. (1985) Manufacturing Strategy: Concepts and Methods. Management of
Productivity and Technology in Manufacturing, P. R. Kleindorfer (ed.), Plenum Press, pp. 153-188.

Cohen, M.A. and Lee, H.L. (1989) Resource Deployment Analysis of Global Manufacturing and Distribution
Networks. Journal of Manufacturing Operations Management, 2, pp. 81-104.

Cohen, M.A. and Lee, H.L. (1988) Strategic Analysis of Integrated Production-Distribution Systems.
Operations Research, pp. 216-228.

Cohen, M. A., Lee, H. L., and Moon, S. (1988) An Integrated Model for Manufacturing and Distribution
Systems Design, Working Paper, Wharton School, University of Pennsylvania, Philadelphia, PA.

Cole, M. H. (1995) Service Considerations and the Design of Strategic Distribution Systems, PhD Thesis,
School of Industrial and Systems Engineering, Georgia Institute of Technology, Atlanta, GA.

Dasu, S. and de la Torre, J. (1997) Optimizing an International Network of Partially Owned Plants Under
Conditions of Trade Liberalization. Management Science, 43 (3), pp. 313-333.

Davis, E. W., (1992) Global Outsourcing: Have U.S. Managers Thrown the Baby Out with the Bath Water?,
Business Horizons, July-August, pp. 58-65.

Ernst, R., Kouvelis, P., and P. P. Dornier (1998) Global Operations and Logistics: Text and Cases, Wiley
and Sons, New York.

Fagan, M. L. (1991) A Guide to Global Sourcing, The Journal of Business Strategy, March-April, pp. 21-25.

Francis, R. L., McGinnis, L. F., and J. A. White (1992) Facility Layout and Location: An Analytical
Approach, Prentice Hall, Englewood Cliffs, NJ.

29
Flipo,C. and Finke, G. (1997) An integrated model for an industrial production-distribution problem,
Working Paper, University of Grenoble, Grenoble, France.

Ganeshan, R, Jack, E., Magazine, M., and P. Stevens, “A Taxonomy Review of Supply Chain Management
Research,” Chapter 27, Quantitative Models for Supply Chain Management, Tayur, S., Ganeshan, R., and M.
Magazine (eds), Kluwer International Series, Kluwer Academic Publishers, Boston, 1999, 839-879.

Geoffrion, A.M. (1977) A Priori Error Bounds for Procurement Commodity Aggregation in Logistics
Planning Models, Naval Research Logistics Quarterly, 24 (2), 201-212.

Geoffrion, A. M. (1974) Multi-commodity Distribution System Design by Benders Decomposition.

Management Science, Vol. 20, pp. 822-844.

Geoffrion, A. M., and R. F. Powers (1995) Twenty Years of Strategic Distribution System Design: An
Evolutionary Perspective, Interfaces, 25, pp. 105-128.

Glasserman, P. and Wang, Y. (1996) Leadtime-Inventory Trade-Offs in Assembly-to-Order Systems.

Working Paper, Columbia Business School, Columbia University, NY, Operations Research, to appear.

Goetschalckx, M., Cole, M.H., Dogan, K., and Wei, R.-P. (1996) A Generic Model for the Strategic Design
of Production-Distribution Systems, Working Paper, School of Industrial and Systems Engineering, Georgia
Institute of Technology, Atlanta, GA.

Graves, S. C., and M. L. Fisher. (1997) Introduction to Special Issue on Frontier Research in Manufacturing
and Logistics, Management Science, 43 (10), pp. 403-404.

Graves, S. C., Kletter, D. B., and W. B. Hetzel. (1998) A Dynamic Model for Requirements Planning with
Application to Supply Chain Optmization, Operations Research, 46 (supp 3), pp. S35-S49.

Hise, R. T., (1995) The Implications of Time Based Competition on International Logistics Strategies,
Business Horizons, JAI Press, Grennwich, CT, 38 (5), Sept./Oct., pp. 39-45.

Hodder, J. E., and M. C. Dincer (1986) A Multifactor Model for International Plant Location and Financing
Under Uncertainty, Computers in Operations Research, 13 (5), pp. 601-609.

Huang,H.-J., Xu, G. (1998) Aggregate scheduling and network solving of multi-stage and multi-item
manufacturing systems, European Journal of Operational Research 105, pp. 52-65.

Kogut, B. (1985) Designing Global Strategies: Profiting from Operational Flexibility, Sloan Management
Review, Fall, pp. 27-38.

Kogut, B. and Kulatilaka, N. (1994) Operating Flexible, Global Manufacturing, and the Option Value of a
Multinational Network. Management Science, 40 (1), pp. 123-139.

Kouvelis, P., and M. J. Rosenblatt (1997) Development of International Facility Networks with Financing,
Trade Tariff and Tax Considerations, Working Paper, John M. Olin School of Business, Washington Univ.,
St. Loius, Mo., December.

Lee, H. L. (1997) Editor’s Introduction, IIE Transactions, 29, p. 626.

30
Lee, H. L. (1998) Global Supply Chain and Technology Management, Vol. 1, POMS Book Series in
Technology and Operations Management, Production and Operations Management Society, PC 543, Florida
International University, Miami, FL.

Lee, H. L., and C. Billington. (1992) Managing Supply Chain Inventory: Pitfalls and Opportunities, Sloan
Management Review, Spring, pp. 65-73.

Lessard, D. R., and J. B. Lightstone (1986) Volatile Exchange Rates Can Put Operations at Risk, Harvard
Business Review, July-August, pp. 107-114.

Little,D., Hemmings,A., (1994) Automated assembly scheduling: A review, Comp. Int. Man. Syst. 7(1), 51-
61.

Luss, H. (1982) Operations Research and Capacity Expansion Problems: A Survey, Operations Research, 30
(5), pp. 907-947.

MacCormack, A. D., Newman III, L. J., and D. B. Rosenfield. (1994) The New Dynamics of Global
Manufacturing Site Location, Sloan Management Review, Summer, pp. 69-80.

Monezka, R. M., and R. J. Trent. (1991) Global Sourcing: A Development Approach, International Journal
of Purchasing and Materials Management, pp. 2-8.

Ohmae, K. (1989) The Global Logic of Strategic Alliances, Harvard Business Review, March-April, pp. 143-
154.

Phillips,C., Stein,C., and J. Wein, (1997). Task Scheduling in Networks, SIAM J. Discrete Math. 10 (4),
pp.573-598.

Rajagopalan, S., Singh, M.R., and Morton, T.E. (1998) Capacity Expansion and Replacement in Growing
Markets with Uncertain Technological Breakthroughts, Management Sciences, 44, (1),12-30.

Revelle, C.S. and Laporte, G. (1996) The Plant Location Problem: New Models and Research Prospects,
Operations Research, 44 (6), 864-874.

Rosling, K. (1989) Optimal Inventory Policies for Assembly Systems Under Random Demands. Operations
Research, 37 (4), pp. 565-579.

Russell,R., Taylor,B. (1985) An evaluation of scheduling policies in a dual resource constrained assembly
job, IIE Transactions, Vol. 17(3), pp. 219-232.

Schmidt, C.P. and Nahmias, S. (1985) Optimal Policy for a Two-Stage Assembly System Under Random
Demand. Operations Research, 33 (5), pp. 1130-1145.

Schraner, E. (1996) Capacity/Inventory Trade-Offs in Assemble-to-Order Systems. Working Paper, IBM T.J.
Watson Research Center, Yorktown Heights, NY.

Taylor, G.D. (1997) Design for Global Manufacturing and Assembly. IIE Transactions, 29, pp. 585-597.

Tayur, S., Ganeshan, R., and M. Magazine (eds), Quantitative Models for Supply Chain Management,”
Kluwer International Series, Kluwer Academic Publishers, Boston, 1999.

Thomas, D., and P. M. Griffin (1995) Coordinated Supply Chain Management: A Review, Working paper,

31
School of Industrial and Systems Engineering, Ga. Institute of Tech., Atlanta, Ga.

Tompkins, J.A., White, J.A., Bozer, Y. A., Frazelle, E. H., Tanchoco, J. M. A., and J. Trevino. (1996)
Facilities Planning, John Wiley and Sons, New York.

Van Roy, T. J. (1983) Cross Decomposition for Mixed Integer Programming, Mathematical Programming,
25, pp. 46-63.

Van Roy, T. J. (1986) A Cross Decomposition Algorithm for Capacitated Facility Location, Operations
Research, 34, pp. 145-163.

Van Roy, T. J., and D. Erlenkotter (1982) A Dual Based Procedure for Dynamic Facility Location,
Management Science, 28, pp. 1091-1105.

Verter, V., and M. C. Dincer (1995a) Facility Location and Capacity Acquisition: An Integrated Approach,
Naval Research Logistics Quarterly, 42, pp. 1141-1160.

Verter, V. and M.C. Dincer (1995b) Global Manufacturing Strategy, Facility Location: A Survey of
Applications and Methods, Z. Drezner (ed), Springer-Verlag, New York.

Verter, V. and Dincer, M.C. (1992) Invited Review – An Integrated Evaluation of Location, Capacity
acquisition, and Technology Selection for Designing Global Manufacturing Strategies. European Journal of
Operational Research, 60, pp. 1-18.

Vidal, C. J. and M. Goetschalckx (1996) The Role and Limitations of Quantitative Techniques in the
Strategic Design of Global Logistics Systems, Ga. Institute of Tech., Atlanta, Ga.

Vidal, C. J., and M. Goetschalckx (1997) Strategic Production – Distribution Models: A Critical Review with
Emphasis on Global Supply Chain Models, European Journal of Operational Research, 98, pp. 1-18.

Vidal, C.J. and Goetschalckx, M. (1998) A Global Supply Chain Model with Transfer Pricing and
Transportation Cost Allocation, Working Paper, School of Industrial and Systems Engineering, Georgia
Institute of Technology, Atlanta, GA.

Wood, D. F., Barone, A., Murphy, P., and D. L. Wardlow (1995) International Logistics, Chapman and Hall,
New York.

Zhang, A.X. (1996) Demand Fulfillment Rates in an Assemble-to-Order System with Multiple Products and
Dependent Demands Working paper, University of Southern California, Los Angeles, CA, pp. 1-31.

32
 K1 K2 K3 S11j

S12j

K4 K5 K6 S21j S13j

S22j

K7 K8 K9

G1 G2

Fig.1: Graphical representation of (a) a logistics network, G1, and (b) a component tree, G2

T11j1

T12j4

T21j2
T13j5

T22j3

Fig.2: An example assembly tree

33
 p11j1

T11j1
q112j14
p12j4

T12j4
q212j24 q123j45
p21j2
p13j5

T21j2
T13j5
q223j35
p22j3

T22j3

Fig.3: Example assembly t

34