Risk-Aware Secure Supply Chain Master Planning

Axel Schr¨opfer

, Florian Kerschbaum

, Dagmar Sadkowiak

and Richard Pibernik

SAP Research Karlsruhe, Germany

Supply Chain Management Institute, EBS Wiesbaden, Germany

Abstract. Supply chain master planning strives for optimally aligned produc-

tion, warehousing and transportation decisions across a multiple number of part-

ners. Its execution in practice is limited by business partners’ reluctance to share

their vital business data. Secure Multi-Party Computation can be used to make

such collaborative computations privacy-preserving by applying cryptographic

techniques. Thus, computation becomes acceptable in practice but for additional

cost of complexity depending on the protection level chosen. Because not all

data to be shared induces the same risk and demand for protection, we assess the

risk of data elements individually and then apply an appropriate protection. This

speeds up the secure computation and enables signiﬁcant improvements in the

supply chain.

1 Introduction

Supply chain master planning (SCMP) strives for optimally aligned production, ware-

housing and transportation decisions across multiple partners. In practice, we can com-

monly observe a decentralized coordination mechanism (referred to as upstream plan-

ning) that usually only leads to local optima rather than to global supply chain optima

[9]. At least in theory, optimal master plans can be generated for the whole supply chain

if some planning unit has at its disposal all relevant information pertinent to the indi-

vidual partners in the supply chain. It is, however, a well known fact that companies are

typically not willing to share sensitive private data (e.g. cost and capacity data) ([17,

18]). They perceive the risk that the central planning unit or other parties misuse data

to their disadvantage in order to obtain additional beneﬁts.

The major obstacles to centralized master planning can be removed if a mechanism

for securely and privately computing the supply master plan is in place [1]. A central

planning unit, e.g. a 4th party logistics provider (4PL), could then determine globally

optimal master plans and distribute these to the individual partners involved in the sup-

ply chain. To this end, Secure (Multi-Party) Computation (SMC) can be employed such

that the relevant data does not need to be disclosed even to central planning unit. This

offers the ultimate level of protection, since no data sharing risk remains. In this paper

we propose a framework for secure centralized supply chain master planning (SSCMP).

We introduce a basic model for centralized supply chain master planning and, from this,

derive the relevant data a central planning unit requires to optimally coordinate man-

ufacturing and transportation decisions. We then analyse this data with respect to its

Schröpfer A., Kerschbaum F., Sadkowiak D. and Pibernik R. (2009).

Risk-Aware Secure Supply Chain Master Planning.

In Proceedings of the 7th International Workshop on Security in Information Systems, pages 46-56

DOI: 10.5220/0002197300460056

 SciTePress

”criticality”. Criticality refers to how sensitive certain pieces of data are and how will-

ing the different partners will be to share this data. The criticality is determined by the

perceived risks associated with data sharing and its prior public knowledge. In this con-

text, risk can be characterized by the potential negative impact that occurs if a partner

misuses the data to its own beneﬁt and the likelihood for this to happen. We derive

an overall criticality assessment for each data element that is relevant for supply chain

master planning and use information about on the prior (public) knowledge of the data

to determine an overall criticality score. This criticality score constitutes an input to se-

cure computation of centralized supply chain master plans. We map criticality scores to

protection levels which consist of certain technologies and parameters for SMC. Lower

protection levels lead to faster SMC implementations. We propose a mixed approach to

SMC combining the different protection levels in one implementation and formulate a

new pivot selection rule in Linear Programming (LP) that optimizes the effort involved

by selecting based on the protection level. We experimentally verify the effectiveness

of the new algorithm.

2 Related Work

Numerous works in the area of supply chain management exists on supply chain master

planning as well as information sharing and collaboration in supply chains. In general, it

is a well acknowledged fact that sharing relevant information and planning in a collab-

orative fashion can improve supply chain performance and mitigate the consequences

of demand variability, especially with respect to the well-known bullwhip effect (see

for example [6, 14,16,23]). With respect to supply chain master planning, numerous

authors have proposed multi-stage models that can be utilized to coordinate planning

activities across multiple locations and ﬁrms (e.g. [10,13, 20]). Various authors have

stated that employing a centralized approach to master planning will lead to better re-

sults as compared to decentralized approaches that are most commonly employed in

industry. [21], for example, analyze the disadvantages of upstream coordination in com-

parison with centralized coordination. They compute the average gap between central-

ized and upstream coordination for several test scenarios with varying cost parameters

and demand patterns. Similar ﬁndings are reported in [18]. However, centralized supply

chain planning has not been widely adopted in industry. [12] states: ”it is difﬁcult, or

maybe even impossible, to get a large network consisting of independent companies

to agree on and implement a centralised planning and control solution.” Reluctance

towards information sharing (a prerequisite for centralized master planning) has been

identiﬁed as the main obstacle that inhibits centralized master planning ([17,18]). For

this reason, alternative approaches have been developed that either build on negotiation

based coordination ([9]) or hybrid forms ([17]). So far there has been no research on

supply chain master planning based on mechanisms that privacy preserving data sharing

and computation. To the best of our knowledge the only approach to secure multi-party

computation in the area of supply chain management can be found in [1]. The authors

develop secure protocols for a Collaborative Planning, Forecasting, and Replenishment

(CPFR) process. Next to the fact that we, in our paper, consider a different problem set-

ting, a major distinction between the research presented in [1] and our research is that

they do not consider different protection levels for different risks of data to be shared.

They follow the approach to provide the highest protection for all data using a spe-

cially developed protocol. Their protocols are two-party protocols, while we consider a

multi-party problem. We will now review related work for SMC.

SMC allows a set of n players, P = P

, . . . , P

, to jointly compute an arbitrary func-

tion of their private inputs, f(x

, . . . , x

). The computation is privacy preserving, i.e.

nothing else is revealed to a player than what is inferable by his private input and the

outcome of the function. A cryptographic protocol is then run between the players in

order to carry out the computation. Even if there are adversarial players, the constraints

on correctness and privacy can be proven to hold under well stated settings. These

settings consider the type of adversary as well as his computing power which can be

bounded or unbounded. An adversary can be passive, i.e. following the protocol cor-

rectly but trying to learn more or he can be active, by arbitrarily deviating. For the

two-party case it has been proven by Yao in [22], that any arbitrary function is com-

putable in privacy preserving fashion, using garbled binary circuits. This approach has

been extended to the multi-party case in [4,11]. Alternative approaches base on secret

sharing schemes. A player’s secret s is split into m shares which are then distributed

to m players. Players can compute intermediate results on the shares, and in the end a

reconstruction is performed in order to receive the ﬁnal result. Other approaches utilize

semantically secure homomorphic encryption (HE) [8], a public encryption scheme,

where E(x) · E(y) = E(x + y) and x cannot be deduced by E(x).

Using the general approach leads to solutions that have high complexity and are there-

fore almost always not practically feasible [15]. Thus, in order to get a practical solu-

tion, a dedicated protocol should be constructed. Atallah et al. constructed solutions for

a couple of supply chain problems, e.g. planning, forecasting, replenishment, bench-

marking, capacity allocation and e-auctions ([1–3]). Their cryptographic protocols base

on additive secret sharing, homomorphic encryption and garbled circuits. A contribu-

tion of Atallah et al. which is closely related to ours is that of secure linear programming

[15]. It uses the simplex method introduced by Dantzig in [7] to solve linear programs

which get expressed as a matrix D. The method consists of two steps: selecting the

pivot element d

and pivoting all elements d

of D over this element. The pivot step

sets the new value of d

, denoted d

′

, by

′

for i = r and j = s (pivot element)

′

for i = r and j 6= s (pivot row)

′

−d

for i 6= r and j = s (pivot column)

′

−d

for i 6= r and j 6= s (all other elements).

The method is repeated until the optimal solution of the LP is found (resp., it is stated

that the problem is unbounded or infeasible). As input to the cryptographic protocol,

matrix D gets additively split between both parties (i.e., D = D

(a)

+ D

(b)

). In order

to not reveal additional information (e.g. by the pivot column or row index), the matrix

gets permuted at the beginning of each iteration. Details are omitted here, but can be

found in [15]. The pivot element selection and the pivot step are then carried out using

cryptographic tools additive splitting, homomorphic encryption and garbled circuits.

3 Supply Chain Master Planning

In this section we ﬁrst provide a basic model for centralized supply chain master plan-

ning. This model will be used to derivethe relevantdata that partners in the supply chain

need to share for centralized master planning. We then propose a simple approach to

assess the criticality of the individual elements.

3.1 Model for Centralized Supply Chain Master Planning

As a basis for our subsequent analysis we utilize a simple generic supply chain master

planning model presented in [18]. Although rather simple, this model is sufﬁcient for

the illustration of our concept and can easily be extended in order to account for further

practical requirements and restrictions.

We consider a supply chain with I stages on which different operations (e.g. manufac-

turing, warehousing, etc.) are performed. We use index i (i = 1, . . . , I) to distinguish

the different stages. By I + 1 we denote the ﬁnal customer stage. By K

we denote

the set of nodes on stage i. Every node k ∈ K

represents one production facility or

warehouse on stage i = 1, . . . , I. The ﬁnal customer locations are modelled through

nodes k ∈ K

I+1

on stage i = I + 1. By N

we denote the set of products produced

on stage i and use m ∈ N

i−1

and n ∈ N

as indices for the input and output products

of stage i. For a given supply chain, master planning determines the production and in-

ventory quantities for every node and the material ﬂows between the nodes for a given

time period. We introduce the following additional notation to formulate a centralized

master planning model:

Master planning parameters (input)

Demand for ﬁnal ﬁnished product n ∈ N

at customer location l ∈ K

I+1

m,n

Quantity of input product m required for manufacturing one unit of output product n

Unit capacity requirement at location k ∈ K

for output of product n ∈ N

cap

i,k

Production capacity at location k ∈ K

i,k

Unit production costs of product n ∈ N

at location k ∈ K

i,k,l

Unit shipping costs of product n ∈ N

from location k ∈ K

to location l ∈ K

i+1

i,k

Unit holding costs of product n ∈ N

at location k ∈ K

Master planning variables (output)

i,k

Production quantity of output product n ∈ N

manufactured at location k ∈ K

i,k,l

Shipping quantity of product n ∈ N

shipped from location k ∈ K

to l ∈ K

i+1

The following deterministic, linear programming model can be used to determine a

supply chain master plan.

Objective function

Min C =

i=1

k∈K

n∈N

i,k

i=1

k∈K

n∈N

i,k

i=1

k∈K

l∈K

i+1

n∈N

i,k,l

(1)

Constraints

k∈K

I,k,l

= D

∀n ∈ N

, l ∈ K

I+1

(2)

i,k

l∈K

i+1

i,k,l

∀n ∈ N

, i ∈ {1, . . . , I}, k ∈ K

(3)

j∈K

−1

i,j,k

n∈N

m,n

i,k

∀m ∈ N

i−1

, i ∈ {1, . . . , I}, k ∈ K

(4)

n∈N

i,k

≤ cap

i,k

∀i ∈ {1, . . . , I}, k ∈ K

(5)

i,k

, y

i,k,l

≥ 0 ∀n ∈ N

, i ∈ {1, . . . , I}, k ∈ K

(6)

The objective of the model is to minimize the total relevant costs of the SC for

fulﬁlling ﬁnal customer demand. The objective function (1) accounts for production

costs, holding costs, and shipping costs for ﬁnished products. Constraints (2) ensure that

the ﬁnal customer demand at stage I + 1 is met. (3) and (4) represent ﬁnished product

and intermediate product balance constraints. The capacity constraints (5) ensure that

the available capacity of any location will not be exceeded. Constraints (6) ensure non-

negativity of all decision variables.

The output of this model is a supply chain master plan for a single period that speciﬁes

the production quantity for the individual products in each node and the shipping quan-

tities across the whole supply chain. From this basic model we can directly infer the

relevant data that needs to be shared in order to realize centralized supply chain master

planning. All parties in the supply chain need to make the above listed input parame-

ters available to the central planning unit. After generating the supply chain master plan,

the central planning unit has to communicate the results (i.e. the values of the output

variables) to the corresponding partners. In typical industry settings, both the input pa-

rameters and the master planning output constitute private data that is only accessible to

the planning units (ﬁrms, departments) responsible for individual nodes and arcs. The

willingness to share this data will depend on the risk the individual data owners per-

ceive. The perceived risk, however, is not identical for all of the relevant data elements.

A company may, for example perceive a low risk associated with sharing forecast data,

but a high risk when revealing production cost or capacity information. While SMC can

overcome the risk of data sharing in theory with the highest protection level, in practice

such solutions can become too slow to be useful (e.g. if the computation takes longer

than what the continuous planning period is). We therefore use the result of the risk

assessment, the criticality scores, to optimize the SMC, such that each data element is

handled at its appropriate risk level. We achieve a signiﬁcant performance improvement

in our experiments.

3.2 Data Criticality and Protection Levels

In this section we illustrate a simple approach to determine protection levels for indi-

vidual data elements in the context of centralized master planning. Although it is rather

straightforward to see that the risk will differ across the individual data elements, it is

not possible to determine general criticality levels that are valid for any supply chain

setting. Whether other partners in the supply chain can use data to their beneﬁt and to

the disadvantage of the data owner depends on factors such as the distribution of power

among the partners, the type of industry and product, the relative position in the supply

chain, trust among partners, etc. Production costs, for example are generally considered

as critical data that a data owner will not want to share. However, in many industries

(e.g. for commodities) production costs are known by different partners without imply-

ing a negative impact. Because a general assessment of the criticality is not likely to be

attainable, an individual assessment has to be conducted for any speciﬁc supply chain.

We propose a simple scheme to support such a criticality assessment. It is based on the

following questions that need to be answered for any one of the data elements identiﬁed

in the previous section:

1. What disadvantage may a data owner potentially incur when sharing private data?

2. What is the probability that a partner in the SC (mis-) uses the shared data to the

disadvantage of the data owner?

3. To what extent is the data prior knowledge?

With the ﬁrst two questions we capture the individual components of the risk in-

duced by sharing a certain data element. When considering the potential negative im-

pacts (question one), we have to consider that these may vary depending on the position

of the data source within the supply chain and the potential incentives other partners

in the SC may have to (mis-) use the data. We differentiate between partners who are

responsible for nodes on the same stage (competitors) and those who are responsible for

nodes on previous or subsequent stages (supplier-buyer-relationships). For each of the

aforementioned cases it is necessary to assess the likelihood of a disadvantage on the

side of the data owner, i.e. the probability that another partner in the supply chain will

actually make use of the knowledge of the data element (question 2). The risk cannot

be considered independent of the prior knowledge about the data. It is reasonable to

assume that the criticality of certain data elements is lower if the data is already acces-

sible for some or all of the partners in the supply chain. Figure 1, a) illustrates our basic

scheme for assessing the criticality of individual data elements.

Potential negative impacts

- induced by competitors: impact * probability score

- induced by suppliers: impact * probability + score

- induced by buyers: impact * probability + score

[0; 5] [0; 5] [0; 25]

Overall risk assessment (Sum of individual scores):

- Public knowledge (pub): score (pub)

- Prior knowledge of specific stages (spe): + score (spe)

[0; 5]

Prior knowledge

(pk)

Prior knowledge weight: (1- pk/10)

Criticality assessment

= Overall risk assessment * prior knowledge weight:

[0; 75]

[0 ; 10]

[0; 1]

[0; 75]

Input for secure computation

1,8

01530456075

Production quantity

Shipping quantity

Variable production costs

Production capacity

Variable holding costs

Demand

Capacity usage rates

Variable shipping costs

Required input/output rates

Data type

very high lowmedium nonehigh

Criticality level of data

a) b)

Fig.1. a) Determination of criticality. b) Criticality levels of different types of data (example).

We propose a scoring range between zero and ﬁve to adequately assess by dis-

crete values the potential negative impact and the expected probability of data misuse.

Through multiplication of both scores, we obtain a particular risk measure for negative

impacts induced by competitors, suppliers, or buyers. Their addition provides a mea-

sure for the overall risk. The overall risk for each data element is then weighted with a

value that expresses the prior knowledge of data. Similarly, a scoring range from zero

to ﬁve is used to measure the degree of public knowledge in general as well as speciﬁc

knowledge of individual SC partners. The sum of both scores measures the level of

prior knowledge. A score of zero indicates that the data is pertinent to the data owner,

while higher scores indicate that the data may anyways be known prior to centralized

master planning. We determine an aggregate weight for the prior knowledge as in order

to derive the overall criticality level. In Figure 1,b) we provide an example of a possible

outcome of our criticality assessment. With our assessment scheme, a criticality score

between zero and 75 is assigned to each data element.

4 Secure Computation

4.1 Protection Levels

The data criticality analysis of section 3.2 shows that different variables in the SCMP

problem have different protection demand. The data criticality scores of the variables

range from zero to 75. We map a data criticality score to protection levels. A protec-

tion level speciﬁes a concrete set of SMC technologies and their parameters for pro-

tecting a variable. These technologies and parameters are: the computational setting

(information-theoretic, cryptographic or best-effort), the cryptographic tools and the

tool parameters. Dependencies among the different parameters of a protection level are

possible, e.g. there cannot be a SMC computation that is information-theoretically se-

cure, but uses homomorphic encryption as a tool. The protection levels are arranged in

order of the effort for an attacker to infer the protected value. Higher protection levels

require more effort and add more additional complexity. Higher criticality scores map

to higher protection levels. See Table 1 as an exemplary speciﬁcation for protection lev-

els which limits the available cryptographic tools to additive splitting and homomorphic

encryption. Table 1 gives concrete examples for ﬁve possible protections levels which

will be used in later experiments.

Table 1. Protection Levels for the 4PL Scenario.

Protection Setting Tools Tool Parameters

very high inf.-theo. additive split w/ modulus modulus N, number of parties

high cryptographic HE Darmgard-Jurik, key: 3072 bit

medium cryptographic HE Paillier, key: 512 bit

low best effort additive split w/o modulus

none - -

4.2 Mapping

A monotone function maps the criticality score c to a protection level p = f(c). We

propose a linear mapping. Other mappings are possible and may depend on the conrecte

application context. We assume the ordered protection levels to differ in their effort by

an almost constant factor. For a ﬁrst mapping we deﬁne a linear mapping function f(c)

which maps data criticality score c to m protection levels by f (c) = 1 + ⌊c · m/(c

max

1)⌋, where c

max

in our case is 75 as received from section 3.2. Applying this mapping

to the criticality scores of section 3.2, we receive the values of Table 2. Considering

Table 2. Linear Mapping.

Protection Level 1 2 3 4 5

Criticality Score 0-15 16-30 31-45 46-60 61-75

Number of Data 3 4 0 2 2

Table 2, nine variables get a protection level assigned below the maximum. Thus, for

nine of eleven variables computational effort can be reduced compared to the former

approach of applying maximum protection.

4.3 Pivot Rule/Intergration

We adapt the solution by Atallah et al. for secure linear programming. We introduce

an additional matrix denoted P . Every element of P , p

, represents the protection

level of the corresponding data element in D, d

. P is available to both parties, as

well as the table with the protection level speciﬁcations. Each party may deﬁne its own

mapping function. Whenever a pivot step is performed in order to receive a new value

′

, the new protection level value p

′

is set to the highest assigned protection level

value of all elements of D involved. According to the pivot step computation rules for

processing the current element d

the involved elements are: d

, d

and d

The new protection level for d

′

then is received by max(p

, p

). Over time,

this leads to convergence of matrix P to the highest protection values contained. We

construct a pivot selection rule which not only bases on entries of D but also on these

of P and moreover prevents P from fast convergence. Recall that the LP is rewritten as

a matrix D. Let

D =



−z

A b



where c

denotes the vector of the objective function’s coefﬁcients, z

the outcome, A

the coefﬁcients of the constraints and b the vector of the constraint values. The secure

linear programming solution originally uses a slight adaption of the Bland’s Rule [5]

as pricing scheme. The rule computes the pivot column s by min(s : c

< 0) and the

pivot row r by min(r : b

) for all a

> 0. Our approach keeps the part of the

Bland’s Rule for selecting the pivot column. We then replace the part for selecting the

row. We deﬁne r for 0 ≤ r ≤ m by

r = min(

) : min



m−1

i=0

n−1

j=0

max(p

, p

) − p



Thus, every element in the selected pivot column s fulﬁlling the minimum ratio test best

(i.e., no row in column s with a smaller ratio exists) is checked for having the lowest

impact on the convergenceof matrix P . Although every element of matrix P is involved

in the computation for every element fulﬁlling the minimum ratio criteria, selecting the

pivot row can still be considered very fast, since P gets updated by each party locally

in the exact same manner and the single operations are less complex. Thus, even for big

m’s and n’s, the added computational overhead can be considered very small. In order

to have the indexes matching, P gets blinded and permuted in the same way as this is

done for D within the original protocol. P may leak little information, e.g. if there is a

unique occurrence of a protection level.

4.4 Experimental Results (Mapping, Pivot Rule)

For further examination, we set up an experiment based on the results of section 4.3 and

the secure linear programming protocol of Atallah et al. For executing the experiment

it is not necessary to actually run the cryptographic protocol, since the main part of

the protocol remains unchanged. We rather focus on the local part, i.e. computing the

elements of the protection level matrix, and run the simplex algorithm locally in order

to simulate pivoting to have correct pricing data available. For preparation of the exper-

iment, we implemented the simplex algorithm in Java, ﬁrst. We then implemented an

instance of a realistic 4PL scenario for medical equipment (details omitted for brevity)

and derived a LP matrix from that which has 191 rows and 480 columns. Table 4 shows

the results of our experiment using the linear mapping introduced in section 4.2. We

receive the total effort by adding the number of assigned protection levels in order to

simulate an execution of the pivot part of the cryptographic protocol. The measured

values are the number of pivot steps, the numbers of steps until convergence and the

total effort.

Table 3. Experimental Results.

Bland’s Rule Modiﬁer Bland’s Rule

Pivot Steps 140 112

Convergence Step 2 99

Total Effort 51249448 32102112

Figure 2 shows the convergence ratio during the run. The convergence ratio is de-

ﬁned as the ratio of the sum of all entries of P divided by the number of elements of P

multiplied with the maximum protection level (i.e., ﬁve).

Fig.2. Convergence Ratio.

The modiﬁcation of the Bland’s Rule led to a decrease of 20% on overall pivot steps.

The protection level matrix was kept from convergence up to step 99 while the conver-

gence ratio quickly reached a value of 0.75. The total effort added by the cryptographic

protocol was reduced to 63%.

5 Conclusions

We introduced a solution for Secure Supply Chain Master Planning (SSCMP) using se-

cure computation. Traditional SCMP computes the optimal production and transporta-

tion plan across a number of parties using Linear Programming. We showed that by risk

assessment and risk handling a signiﬁcant performance increase in SSCMP is possible.

We derived a methodology for risk assessment, the criticality score, in supply chains

and then modify the pricing scheme of Linear Programming handling each data item at

the appropriate risk level. In an experimental study based on a realistic scenario using

this methodology we obtained a performance gain of 37%. Future work is to extend the

applicability of the method to other algorithms for linear optimization, e.g. inner point

methods, and to extend it to other supply chain optimization problems adapting the risk

assessment step.

References

1. M. Atallah, M. Blanton, V. Deshpande, K. Frikken, J. Li, and L. Schwarz. Secure Collabo-

rative Planning, Forecasting, and Replenishment. Working Paper, Purdue University, 2005.

2. M. Atallah, M. Bykova, J. Li, K. Frikken, and M. Topkara. Private Collaborative Forecasting

and Benchmarking. Workshop on Privacy in the Electronic Society (WPES), 2004.

3. M. Atallah, H. Elmongui, V. Deshpande, and L. Schwarz. Secure Supply-Chain Protocols.

Proceedings of the IEEE International Conference on E-Commerce (CEC’03), 2003.

4. D. Beaver, S. Micali, and P. Rogaway. The round complexity of secure protocols. In Pro-

ceedings of 22nd STOC, 1990.

5. R. Bland. New ﬁnite pivoting rules for the simplex method. Math. of Op. Res. 2, 1977.

6. F. Chen, Z. Drezner, J. Ryan, and D. Simchi-Levi. The bullwhip-effect :managerial insights

on the impact of forecasting and information on variability in a supply chain, in: Taylor, S.,

Ganeshan, R., and Magazine, M. (Eds.), Quantitative Models for Supply Chain Management,

Boston 1999.

7. G. Dantzig, and M. Thapa. Linear Programming 1: Introduction. Springer-Verlag, 1997.

8. I. Damg˚ard, and M. Jurik. A generalisation, a simpliﬁcation and some applications of pail-

lier’s probabilistic public-key system. In International Workshop on Practice and Theory in

Public Key Cryptography (PKC) 2001, 2001.

9. G. Dudek, and H. Stadtler. Negotiation-based collaborative planning between supply chain

partners, in: European Journal of Operational Research 163, 2005.

10. B. Fleischmann, and H. Meyr. Planning Hierarchy, Modeling and Advanced Planning Sys-

tems, in: De Kok, A. G., Graves, S. C. (Eds.): Supply Chain Management: Design, Coor-

dination and Operation, Handbooks in Operations Research and Management Science, Vol.

11, Amsterdam 2003.

11. O. Goldreich, S. Micali, and A. Wigderson. How to play any mental game. In Proceedings

of the 19th annual ACM symposium on Theory of computing, 1987.

12. J. Holstr¨om, K. Fr¨amling, J. Tuomi, M. Krkkinen, and T. Ala-Risku. Implementing collabo-

ration process networks, in: The International Journal of Logistics Management 13(2), 2002.

13. V. Jayaraman, and H. Pirkul. Planning and coordination of production and distribution facil-

ities for multiple commodities, in: European Journal of Operational research, Vol. 133.

14. H. Lee, V. Padmanabhan, and S. Whang. Information distortion in a supply chain: the bull-

whip effect, in: Management Science, Vol. 43, No. 4, 1997.

15. J. Li, and M. Atallah - Secure and Private Collabortive Linear Programming. 2nd Interna-

tional Conference on Collaborative Computing: Networking, Applications and Worksharing

(CollaborateCom), 2006.

16. S. Min, A. Roath, P. Daugherty, S. Genchev, H. Chen, A. Arndt, and R. Glenn Richey. Supply

chain collaboration: what’s happening?, in: The International Journal of Logistics Manage-

ment, Vol. 16, No. 2, 2005.

17. R. Pibernik, and E. Sucky. Centralised and decentralised supply chain planning, in: Interna-

tional Journal of Integrated Supply Management 2(1/2), 2006.

18. R. Pibernik, and E. Sucky. An approach to inter-domain master planning in supply chains in:

International Journal of Production Economics V. 108, 2007.

19. A. Shamir. How to share a secret. Communications of the ACM, 1979.

20. J. Shapiro. Modeling the Supply Chain, Paciﬁc Grove 2001.

21. N. Simpson, and S. Ereng¨uc. Modelling the order picking function in supply chain systems:

formulation, experimentation, and insights, in: IIE Transaction 33(2), 2001.

22. A. Yao. Protocols for secure computations. In Proc. 23rd IEEE Symposium on the Founda-

tions of Computer Science (FOCS), IEEE, 1982.

23. Z. Yu, H. Yan, and T. Cheng. Beneﬁts of information sharing with supply chain management,

in: Industrial Management and Data Systems, 2001.