Optimizing dynamic supply chain formation in

Inf Syst Front DOI 10.1007/s10796-012-9380-y

Optimizing dynamic supply chain formation in supply mesh using CSET model Hang Yang & Simon Fong

# Springer Science+Business Media, LLC 2012

Abstract A new e-Service model called dynamic supply chain is characterized by their dynamic nature in easily being formed and disbanded with the seamless connectivity provided by e-Marketplace. The new term “supply mesh” was coined to represent this virtual community of companies in which dynamic supply chains, as per project (also known as make-toorder), are formed across different tiers of suppliers. In a supply mesh, a dynamic supply chain can be formed vertically, from the top to the bottom layers, mediating different companies for a project. Companies that are on the same level laterally are usually competitors, and the companies that are linked vertically as supply chains are trading partners. From a global view, the companies that are connected in the supply mesh can be viewed as individual entities that have self-interest. They may compete for survival as well as collaborate with each other for jobs. Given such complex relations the challenge is to find an optimal group of members for a dynamic supply chain in the supply mesh. A multi-agent model called the collaborative single machine earliness/tardiness (CSET) model was recently proposed for the optimal formation of make-to-order supply chains. This paper investigates the possibilities of applying CSET in a supply mesh, and the corresponding allocation schemes are experimentally studied in simulations. One scheme called Cost-driven principle leads to destructive competition while the other one namely Pareto-optimal evolves into a cooperative competition that tries to mutually benefit every participant. The results, based on samples from the U.S. textile industry, show that a cooperative competition scheme is H. Yang : S. Fong (*) Faculty of Science and Technology, University of Macau, Av. Padre Tomás Pereira, Taipa, Macau SAR e-mail: [email protected] S. Fong e-mail: [email protected]

superior in terms of optimal allocation, which obtains maximum satisfaction for all participants. Keywords Supply chain formation . Pareto-optimization . Simulation . e-Marketplace

1 Introduction The increasing number of e-Services available on the Internet has led to the creation of online e-Marketplaces. The interaction within an e-Marketplace can be viewed as a model of an electronic central hub uniting various entities from different industries with seamless connectivity. Buyers and sellers meet and trade online; logical supply chains can be potentially formed according to their trading needs and the product or services that they offer. Software agents running at the central hub serve as mediators that match potential buyers and sellers in the e-Marketplace. Figure 1 shows an example of an electronic hub that connects to different companies via agents. Supply chain formation can be viewed as a mechanism of determining the production and exchange relationships across a number of companies engaged in a coalition. Traditionally, supply chains were formed and maintained over long periods of time through extensive human negotiation for agreeable terms, typically involving a small number of companies at a time (Harland 1996). However in an e-Marketplace companies can easily and quickly reach out and stay connected to a large virtual community, has made the supply chain formation process more dynamic. The hub model which is used to mainly provide connection services in the past evolved to the problem of dynamic supply chain formation with the aim of maximizing mutual benefits among the trading partners over the technical infrastructure.

Inf Syst Front

Fig. 1 Agent-mediated electronic hub as an e-Marketplace

Dynamic supply chain formation ensures that business interactions can be quickly and flexibly formed and easily dissolved to better respond to rapidly changing market conditions. As a result, some built-to-order or make-to-order supply chain management models have emerged. Make-to-order is a contemporary production approach in which a confirmed order for a product is received, and then the product is tasked to be built. This approach is widely used for highly configured products (Holweg and Pil 2004). These dynamic supply chains are often short-lived–quickly formed per project and then dispersed when the service is no longer needed. They are characterized by highspeed, automated mediation aided by software agents (Lo and Kersten 1999; Buyya et al. 2000) that are used to mediate the formation and subsequent coordination of supply chains. Techniques and protocols that enable dynamic supply chain formation coalitions among agents have been widely proposed in the literature. However, they are typically focused on the technical aspects of the formation (e.g., system integration, communication protocols, etc.) and do not address the big picture of how a supply chain will operate. Thus, a high-level view is desired that shows how a dynamic supply chain is formed by selecting the right candidates from all the available parties connected in an eMarketplace. This inclusive view is important because it helps supply chain managers to visualize the anticipated outcome of the formation from a decision-making/support perspective before an actual commitment is made. A supply mesh is a new way of conceptualizing complex, inter-dependent, loosely-coupled and demand-driven business relations knitted within an e-Marketplace. In real-life we have examples of supply mesh consortiums that only approve the participation of certain (usually qualified) companies with the potential to mutually benefit one other. Logically, it functions like a conceptual relation map that depicts the links between

supply chain participants across several tiers, with multiple supply chains as conceptual business corporations/partnerships among these participants functioning simultaneously. When viewed from the perspective of a single supply mesh, the selection problem is similar to a combination of resource allocation and resource scheduling problems, determining who the participants in the supply chain are, each participant’s share of the jobs and the timely sequence of job executions among the participants. For example, in Fig. 2a there are three conceptual supply chains being formed through retailers, manufacturers and material suppliers that are electronically connected in an e-Marketplace; however, their demand/supply relationships as supply chains are reflected on a supply mesh. Each supply chain exists to execute a particular project, and the relationship only lasts the project’s lifetime. The distribution of resources crosses all partners in the supply chain during the lifetime of a project, including retailers, tiers of manufacturers and raw material suppliers. The corresponding view of resource allocation of a supply mesh is shown in Fig. 2b. Before forming a dynamic supply chain on a supply mesh, decisions must be made regarding the selection of the right participants to support and fulfill the production of the product. Supply chains are formed on a supply mesh using two general approaches: (global) computation-based and (local) negotiation-based. A computation-based approach extensively considers all the complex factors from a global view to find a solution that best satisfies high-level goals. In this case, a high-level central authority finds an optimal group of participants for a particular project in a supply mesh and instructs them to collaborate. The Pareto-optimal principle is usually applied in this method to ensure the majority of the participants’ gains without jeopardizing anyone. Following this principle allows jobs to be shared among multiple winners, which encourages cooperative competition. Cooperative competition is based on promoting mutual survival–an “everyone wins” mentality. It is a process where individuals compete in a cooperative manner through peaceful exchanges and without violating others to improve their level of satisfaction. Cooperative competition focuses entities against the environment (Kropotkin 2005) and jobs are shared as much as possible. A negotiation-based approach assumes an environment in which the companies across each level of a supply chain compete for jobs through private negotiations or bidding efforts. The supply chain formation decisions in this approach are made individually by the companies using local knowledge. This negotiation-based method requires that each agent serve as a mediator to coordinate negotiation among bidders at each level and find the best deals. This competitive approach is usually cost-driven and results in few competitive winners who acquire the majority of the jobs and can almost always achieve the lowest cost in each production project. The supply chains that result from this approach usually cover a few elite industrial players who absorb the majority of the jobs and charge low prices.

Inf Syst Front Fig. 2 Examples of supply chain formation in a supply mesh: left, (a) logical supply chain groups. Right, (b) agentmediated supply mesh with a view of resource allocation

Many researchers have applied Pareto-optimal resource allocation to supply chain management, but there has been no systematic visual comparison of these two different principles. This paper demonstrates how the two competition schemes, namely Pareto-optimal and cost-driven, affect participants in a supply mesh–especially within a long-term view. A Pareto-optimal scheme is used in the computationapproach and a cost-driven scheme is applied to the negotiation-approach for supply chain formation. In particular, we observe by simulating a supply mesh in two modes for the different competition schemes to examine the levels of balance and fairness under competition. The competition relationship in the supply mesh is presented through the visualization output of our simulator. The competition schemes reflect a trend of evolution. The experiment is then extended to find the optimal allocation of jobs as represented by the formation of dynamic chains given some samples taken from a U.S. textile industry. The results show that a cooperative competition scheme is superior in terms of optimal allocations, and that a group of optimal supply chains can be heuristically selected from a supply mesh. The remainder of this paper is organized as follows. A literature review of the relevant topics covered in this paper is in Section 2. Our proposed CSET model for optimizing supply chain formation is presented in Section 3. Experiments for the proposed Pareto-optimal model and the corresponding supply chain formation schemes are shown in Section 5. In particular, we compare three types of Pareto operations: CSET Pareto model versus non-Pareto, CSET Pareto versus 80/20 Pareto, and an extreme case namely co-operative versus destructive competition. Section 8 concludes the paper.

2 Related work To embrace quick changes and fluctuations in market demands, decisions about dynamic supply chain formation usually rely on demand forecasting and negotiation techniques. The former is a preemptive approach that often takes proactive action in resource planning, although forecasting based on historical data

is known to be somewhat unreliable and this approach is sometimes inadequate to offset the surges in supply and demand brought on by pricing adjustments (Banker 2005). The latter approach, implemented through intelligent agent negotiation, is reactive. With no regard for past data, negotiations in this approach ensure the optimal matching of participants by forming a dynamic supply chain–giving its best effort during any changes in market conditions. Many researchers (Holsapple et al. 1995, 1997; Rangaswamy and Shell 1997) have recently favored this approach of dynamic supply chain formation through automated negotiation in contrast to forecasting. Lim and Benbasat (1993) proposed a theoretical model for a negotiation system that is composed of two major components: a decision aid component and an electronic communications component. As the result of the information processing capability of the decision aid component, solutions with automated negotiation displayed better performance than those without (Li et al. 2004). The concept of a web-based intelligent agent system is not entirely new. As early as 1996, Chavez and Maes (1996) developed an agent that assisted users with the exchange of goods through negotiation capability in an emarketplace. This model proactively sought out potential buyers or sellers and automatically negotiated with them on behalf of the users. This model seeded a flexible multi-agent framework for multilateral negotiation. However, these agents were somewhat ‘unguided’ because their designs were general and primitive regarding direct buying and selling, and the negotiation was based only on price. An optimized resource allocation was proposed for mobile device in grid environment (Li and Li 2012). The policy maximized utility of mobile system taking account into energy, budget and deadline. From the perspective of a supply mesh, which maps the logical supply chains within an e-Marketplace, the participants in multiple supply chains must serve in certain roles. The relationship between any two tiers is usually collaborative because one tier needs the other to supply the supplies and fulfill the demands. The relationships between participants in the same tier could be competitive because they are fighting for jobs. This kind of workflow, when viewed as a whole on a supply mesh,

Inf Syst Front

should be optimized in a way that is similar to resource allocations that coordinate across the different relationships and intelligent agents in the different supply chains (Nam 2003; You and Grossmann 2007). A parameterized model was proposed to capture and adjust the information sharing in a supply chain. This model creates new supply chain that can be configured in terms of events or changes in a timely manner (Liu and Kumar 2011) In addition to buy/sell, collaborative and competitive relationships should also be considered. Tian et al. (2006) developed an extended contract net mechanism for supply chain formation in a semi-governed scenario. Their negotiation protocols worked on a semi-monopolized vertical market, China petroleum. Homburg and Schneeweiss (2000) extended the negotiation structure to cover two main features, the strategic changing of the supplier’s production facility and the uncertainty of the retailer’s future demand. The negotiation takes place with respect to different demand scenarios, yielding a variety of possibilities for avoiding negotiation deadlocks. However, both above-mentioned studies by Tian et al. (2006) and Homburg et al. (2006) were based on long-term production supply chains. In other words, all of the production capabilities and transportation capabilities in their studies were based on long-term scheduling. Thus, both their agent systems are only suitable for optimizing mid/long-term, contract-based dynamic supply chain formation rather than for short-term dynamic supply chains. Ideally, a supply mesh should be equipped with negotiation agents that are capable of forming tightly-coupled internal supply chains, each fitting into a suitable role within the overall supply chain production. It should also have the flexibility to solicit qualified participants in an open e-marketplace. Such a supply mesh would be able to react to the ripple effects (Walker 2004) of major economic changes, including the impact of new technology, sudden terrorist/natural disaster threats and policy reforms. Kim et al. (2006) proposed a supply chain model with an important scheduling mechanism for forming an optimal supply chain by means of negotiation and passing information along the supply chain. Inspired by the goals of scheduling optimal resources within a single entity, the mechanism was called the Single machine earliness/tardiness (SET) model, and its scheduling takes both early (earliness) and late (tardiness) production costs into consideration, along with the competitive relationships between multiple participants. The primary theoretical fundamental used by the SET model is Paretooptimality (Fudenberg and Tirole 1983). In line with the Pareto-optimal algorithm, each participant along the dynamic supply chain cannot suffer a loss, meaning that every participant gains from the jobs to varying extents. One potential drawback of the SET model is the relatively long time required for the transfer of cost information and negotiation as the result of the inter-tier connections. To address this, Yang et al. (2010) extended SET to create CSET, which has a central agent to coordinate the negotiation agents on each tier of the supply mesh. Using this double-agent architecture increases efficiency

because the amount of information flow is reduced between each pair of tiers. More importantly, the CSET model functions as an overall scheduler where just-in-time (JIT) mechanisms can be implemented to globally optimize the overall performance of the supply chains. Though the initial concept was described in this paper, this paper reports simulation experiments in details. The highlights of this paper include: (1) A general framework of CSET was defined, but its details such as the workflow and processes in each step are reported here; (2) In this paper a CSET model under destructive competition (non-Pareto) and Pareto was evaluated; (3) CSET model and its mechanisms are verified via extensive experiments for evaluating its characteristics from different perspectives. The well-known JIT principle of supply chain management was initially developed in Japan (Monden 1981). JIT is a production process approach that strives to eliminate sources of manufacturing waste by producing the right part in the right place at the right time so that the ROI improves as the result of reduced inventory levels and delivery lead times. The performance of a zero inventory simulation model for a JIT manufacturing and production system was studied and proven. The main benefits of JIT, such as inventory reduction, quality improvement and quick delivery, are well known (Billesbach 1991; Hobbs 1994; Temponi and Pandya 1995; Cook and Rogowski 1996). Therefore, it makes perfect sense to incorporate JIT principles into CSET to achieve optimal resource allocation and efficiency in dynamic supply chains. CSET lays out a suitable agent operational framework for dynamic supply chain formation. This paper aims to validate this hypothesis via experimental simulations. Furthermore, which resource allocation schemes are chosen has significance for a supply chain’s efficiency. For example, Lau and Zhang (2004) proposed a two-level framework for coalition formation. This framework used a centralized optimization model on the upper level and a distributed agent-negotiation model on the lower level–similar to our CSET model, except that it only had two levels. Hoong and Lei proved that this two-level framework more effectively encouraged the agents to be partially cooperative rather than either fully cooperative or self-interested during the operation of the supply chain. In their subsequent research (Lau et al. 2005), the open constraint optimization problem (OCOP) was solved using this two-level framework. The objective of the generalized OCOP was to find a solution with a low total cost and high overall satisfaction level for agents. On the one hand, their proposals were validated as a set of problem-driven mathematical equations that had not been referred to as a practical application, especially in tests conducted under supply chain management. Supply chain negotiation, however, is more complex than the OCOP problem because it involves special factors such as sequent times, meeting deadlines, penalties and costs. The two-level framework only considered the cost and global satisfaction. On the other hand, every agent

Inf Syst Front

is seen as an autonomy that uses the specified local schemes to define satisfactions. Essentially, the agents on different tiers might have different schemes. Under the competitive condition, intelligent agents with the Pareto-optimal algorithm support negotiation by searching for an agreement that benefits every participant as much as possible; at the same time the central agent in CSET applies the JIT principle to improve the ROI for the whole supply chain. Our proposed automated negotiation model integrates these two methods (Yang et al. 2008). For this reason, our model is able to shorten the scheduling and waiting time so that the dynamic formation of supply chains in a supply mesh can be made more efficient and the overall ROI can be improved.

3 CSET model for optimizing supply chain formation 3.1 CSET model framework The collaborative single machine earliness/tardiness (CSET) model is an extension of the SET model (Kim et al. 2006). It is designed as a methodology for finding an optimal supply chain formation that ensures that every entity in a supply mesh has a job. First, the participating companies that are connected to an e-marketplace would have to join a supply mesh—a logical group of companies that are willing to take part in supply chain collaboration. The CSET model then functions as a multi-agent system governing the supply mesh. This multi-agent system is mainly composed of two functional agents: the Pareto agent (PA) and the JIT collaborative agent (JIT-CA). As shown in Fig. 3, between every two tiers in a supply mesh there is one PA, which runs Pareto-optimal Retailers

Retailers 1

1

8

MA

PA

MA

PA

Suppliers Agent-based SET model

5

6

4

5

JIT-CA

Manufacturers

4

3

6

Material flow

Manufacturers

3

2

7

2

Suppliers CSET model

The Participants: including Retailers, Manufacturers, Supplie The Agents: Mediation Agent (MA), Pareto Agent (PA), Collaborative Agents (CA) Simplex information flow Duplex information flow

Fig. 3 SET model and CSET model framework

computations to optimize the satisfaction of all the participants between the two tiers. All of the PAs work under the JIT-CA, which acts as a central authority. The JIT-CA is then responsible for optimizing the whole supply chain by synchronizing the job allocations. It is assumed that all participants are electronically connected to each other within the emarketplace, and that they are members of a common supply mesh that has one JIT-CA overseeing the supply chain flows: In the CSET model, when a job is initialized by the clients who are usually retailers in the supply mesh, it is first sent to the JIT-CA, who works as an inquisitor to validate the feasibility of the job. Consideration factors include, but are not limited to, participant availability within the supply mesh, profit margins, technical requirements and economic and political issues. The JIT-CA serves as a gate-keeper and administrator of the supply mesh. Once the jobs have been checked and are deemed worthy, they will be represented as job requests and will naturally descend all the way down to be upstream of the supply chain where the raw material suppliers are. The PA serves as the mediator in each layer (section of the supply chain), deciding who gets what shares of each job. In the CSET model, the decision would be made collaboratively between the JIT-CA and the PAs. The JIT-CA monitors the conditions and the wellbeing of each participant, and because it has all the information about the whole supply chain and the supply mesh, the JIT-CA is in a position to derive a global optimal decision based on an overall view. In this scenario it is assumed that the participating businesses in the supply mesh trust the JIT-CA and actively delegate to them all decisions about whether job orders are accepted. Hence, the JIT-CA has the over-riding authority to decide the arrangements over the PAs. The JIT-CA’s main task is to collaborate with all the PAs. Each PA is tasked with submitting information to the JIT-CA, at which point the PA is instructed to adjust its local Pareto decision regarding the job allocations. The old SET model propagates job information through tiers of PAs, but it lacks a global decision-maker to optimize final decisions. In addition, because the SET model is lacking this common channel to the central agent, the amount of information flowing between the PAs and participants is greater. In the SET model, the sum of the information that flows between the participants can be expressed as follows: The sum of the information that flows between the participants for SET and CSET models are expressed respectively as follows: SET : ½½infoFlow## total P  P " " ¼ 2ð # ði ¼ 1Þ ðn  1Þ  m# i þ # ðj ¼ 2Þ n  m# j CSET : infoFlow#total ¼

n1 X i¼1

mi þ

n X

mj þ n  1

j¼2

where n 0 the number of tiers, n≥0 and mn is the number of m participants in tier n, m≥0.

Inf Syst Front

3.2 Operation of the CSET model CSET is the pioneer model that adopted the JIT principle and extended the framework of SET into a collaborative environment where a central agent maintains an efficient workflow and the fairness of job allocation. The CSET model achieves an efficient flow of supply chains by operating tasks in synchronization and reducing the waiting time of the messages as they pass from tier to tier. That means that a pipelining mechanism is facilitated in CSET to ensure that the companies downstream start receiving jobs as soon as the previous job requests have been sent upstream. While staying in communication with each PA, the JIT-CA is able to monitor the overall conditions and remain informed of all happenings in real-time along the supply chains. The JITCA is responsible for reconstructing common requests into sequent requests, namely equilibrium request units by sequence time with job requests being given a time rule. The SET model works within “tardiness” and “earliness” time principles. Combining the PAs and the JIT-CA makes the overall Paretooptimal efficiency higher than when using PAs alone. CSET is designed to work with a generic number of multiple mid-tiers in a supply mesh. For the sake of simple illustration, we use a three-tiered supply mesh. The three tiers are Clients’ Orders (Downstream), Manufacturers (Midstream) and Suppliers (Upstream). Figure 4 explains the operation of the CSET framework with a flow chart that represents a Fig. 4 An example of workflow for dynamic supply chain formation in the CSET model

simple, three-tier supply chain formation. There are n units of orders to be allocated to m number of manufacturers, and the manufacturers require supplies from x number of suppliers. The operational flows between the PAs and the JIT-CA are synchronous because they operate as if under a single entity. The other operations involving companies that are members of the supply mesh connected via the e-marketplace are asynchronous. However, they are given deadlines to respond to requests. In general, the workflow can be partitioned into 9 steps. The works within any one step are atomic and the steps may take place in a sequential order. Step 1: Orders from downstream arise and clients initiate requests to form potential supply chains for producing products. The orders are sent to PA1 for estimation. Step 2: PA1 passes the order requests to the JIT-CA for consideration; the orders are assessed by attempting to reconstruct them into JIT requests. Step 3: JIT-CA returns the results of the order reconstruction (Sequent Request) to PA1 if okay. Negative acknowledgement to PA1 otherwise. Step 4: PA1 distributes those sequent requests to midstream factories. Step 5: According to the orders’ sequent requests, every factory makes a new request for estimation through PA2. Step 6: PA2 sends factories’ requests to the JIT-CA for JIT request validation and reconstruction.

Inf Syst Front

Step 7: JIT-CA reconstructs factories’ requests and returns the sequent requests to PA2; meanwhile, the factories requests are sent to PA1 by the JIT-CA. Step 8: PA2 sends the factories’ sequent requests to all the upstream raw material suppliers and the suppliers propose their scheduling plans and return them to PA2; meanwhile, PA1 makes the Pareto-optimal resource allocation between orders and factories. Step 9: PA2 makes a Pareto-optimal resource allocation between factories and orders. The JIT function will try to shorten the time difference between job requests as much as possible with the aim of pipelining the supply chain’s entire production flow. 3.3 Participants’ input parameters There are four types of participants in a CSET system, and each of them has specified parameters data that must be inputted into the agents for computation to obtain an optimal resource allocation. These input parameters are summarized in Fig. 6 and briefly described as follows: Clients are the order request senders. The data package of job requests sent from the client includes the following information: Amount: This is the quantity of final product that the client proposes to purchase. Location: This is the client’s location, and is used to calculate 3PL delivery costs. Anticipant Delivery Time: This is the expected due-date when the final product will be delivered. Before this date, the client refuses to receive the goods from upstream because early delivery may imply unnecessary storage costs. Latest Acceptable Time: This is the latest due-date at which the client is willing to accept the goods. However, if the delivery comes later than the anticipant delivery time the manufacturer pays the penalty. In other words, the delivery date between the anticipant delivery time and the latest acceptable time is the available zone; otherwise, the delivery is rejected. If so, the manufacturer suffers the loss on its own. Daily Penalty: This is the per-day penalty for late delivery. Price: This is the final product procurement price. Manufacturer is the factory that implements the manufacturing work. It is midstream between the downstream

clients and upstream suppliers. The data package sent from the factory includes the following information: Final Product Amount: This is the amount of the outbound final product that the factory is able to provide. Raw Material Amount: This is the amount of the inbound raw material that the factory needs to fulfill manufacturing. Manufacturing Ratio: This is the ratio of final products and raw materials amounts. For example, one unit amount of the final product is manufactured from two units of raw material A and one unit of raw material B for a manufacturing ratio of Final Product: Raw Material A: Raw Material B01:2:1. Location: This is the location of the factory, used to calculate 3PL delivery costs. Working Duration: This is the period of work time spent in manufacturing. Daily Manufacturing Cost: This is the per-day manufacturing cost that a factory must pay. Daily Inventory Cost: This is the per-day cost of inventory that the factory has to pay to store the final product. The Supplier is accountable for providing raw materials to the manufacturing factories. The data package sent from the supplier includes the following information: Max Supply Amount: This is the maximum amount of raw materials supply to be provided to the manufacturer. Supply Time: This is date on which the supplier is able to provide the supplement. Location: This is the supplier’s location, used to calculate 3PL delivery costs. Price: This is the price of raw material supply. The Third Party Logistics (3PL) Provider is accountable for providing the freight among clients, factories and suppliers. The data package sent from the 3PL provider includes the following information: Delivery Price (per unit weight plus unit distance): This is the delivery price calculated using the distance and weights of the goods; Delivery Time (per unit distance): This is the delivery time calculated using the distance between the starting point and the destination.

Inf Syst Front

3.4 Formulation for CSET model

fPenaltyCost ¼

In order to abstract the operations of job competition schemes in a supply mesh, the following assumptions are made both on business and technical view. The trading contract between manufacturer and suppliers is Free-on-broad (FOB), which means the freight cost is shouldered by the manufacturer. The one between manufacturer and retailer is Cost-and-Freight (CFR), which means the freight cost is also shouldered by the manufacturer. In other words, the manufacturer is responsible for paying all the delivery bills. A supply mesh runs on a make-to-order workflow. The production approach is that, when an order for products is sent, products manufacturing and raw material procuring are initiated. For job allocation result, each type of entity has its specific type of satisfaction, which is defined as the Local Scheme Satisfaction (LSS). It influences the entity’s positive value in supply chain business. The higher the LSS, the higher its positive value in supply mesh is. Data transactions run on a trusted network and with well-defined agent privacy, so that the supply mesh is assumed to be safe from security threats. Generally, there are five kinds of cost involved in our experimental simulation: Fixed Cost, Manufacturing Cost, Inventory Cost, Penalty Cost and Freight Cost. The sum of them is the total cost. Fixed Cost is the cost for raw material procuring. fFixedCost ¼

x X

ðPricei  Amounti Þi

ð1Þ

m X

ðDailyPenaltyCostm  PenaltyDaym Þ ð4Þ

j¼1

Freight Cost is the cost that the manufacturer shall shoulder itself. The cost includes that from supplier to manufacturer and manufacturer to retailer. fFreightCost ¼

m X

ðAmountm  Distancem  UnitCostm Þ

j¼1

ð5Þ

As Fig. 5 shows, the relevant cost calculation above is unified in both experimental supply mesh simulation; however their resource allocating principles are distinct. In terms of the different principles, the supply chains formation in supply mesh results in a disparate outcome as is shown below. 3.5 Cooperative competition agent-mediated supply mesh (CC-ASCE) Destructive Competition Agent-mediated Supply Mesh (DC-ASCE) (Putten et al. 2006) runs the negotiation seeking the lowest cost. It is also called cost-driven supply chain. DC-ASCE resource allocation selects the participants with the lowest overall cost. Cost-driven principle represents the bias that buyers always seek the lowest price with the maximum profit as the resource allocation (Smith 1950). The supply chain mediator-agent distributes the resource by using the lowest cost principle:

i¼1

where x is the total number of suppliers. Manufacturing Cost is the cost to process final products from the raw materials. fManufCost ¼

m  X

Durationj  DailyCostj



ð2Þ

j¼1

where m is the total number of manufacturers. Inventory Cost shall be paid in the case that, if the retailer delivers the final product earlier than the client’s anticipant delivery time, and the delivery is refused by the client. Thus the retailer must suffer the inventory cost in this duration. fInventoryCost ¼

m X

ðDailyCostm  StorageDaym Þ

ð3Þ

j¼1

Penalty Cost shall be paid in the case that, if the retailer’s delivery time is later than the client’s anticipated delivery time while no later than the latest acceptable time, the product is accepted by the client but with some penalty. Likewise, the retailer’s delivery time is evaluated by factory finishing manufacture time and 3PL delivery time.

Min

p P

ðFixedCostn þ ManufCostn þ InventoryCostn

n¼1

ð6Þ

þ PenaltyCostn þ FreightCostn Þ where n is the number of resource allocation. Cooperative Competition Agent-mediated Supply Mesh (CC-ASCE) (Wigand and Benjamin 1995) runs according to the Pareto-optimal principle for resource allocation. In the light of the fairness principle, no one will suffer a loss. The negotiating runs on the basis of weak Pareto-optimal theory. It means a resource allocation can gain as long as at least one allocation has better performance while all other participants are willing to accept it. This better allocation is resulted from weak Pareto-optimal improvement. The performance is justified by the Pareto-optimal principle. A Global Scheme Satisfaction (GSS) is derived from the total cost calculation, which also reflects the supply chain total productivity. The valid range is [0, +100]. The Utility Function of GSS is: UGSS ðCostx Þ   CostMax Costx ¼ 100  Cost ; where UGSS ðCostx Þ 2 ½0; þ100 Max CostMin ð7Þ

Inf Syst Front Fig. 5 The flowchart of two resource allocating principles: destructive competition (nonPareto) and constructive competition (Pareto)

The resource allocating principle in CC-ASCE not only considers the total productivity but also each entity’s satisfaction. For different types of participants, their satisfaction is specified in terms of different factors. We define these factors as three Local Scheme Satisfactions (LSS) which are the selfinterests of the individual from their own (local) view. They are: –

The Distributor’s Scheme Satisfaction (DSS);

UDSS ðDeliveryTimex Þ ¼ 100 

– –

The Manufacturer’s Scheme Satisfaction (MSS); The Supplier’s Scheme Satisfaction (SSS).

DSS is calculated by the time factors. The DSS utility value ranges from −100 to +100; however only the positive values are the distributor’s satisfaction value. The DSS utility function is:

LatestAcceptableDeliveryTime  DeliveryTimex ; where UDSS ðDeliveryTimex Þ 2 ½0; þ100 LatestAcceptableDeliveryTime  AnticipatedDeliveryTime

ð8Þ

SSS can be deduced as the supplying amount. The larger raw material amounts supplier offers, the higher profit he can gain. For this reason, the SSS utility function is: SupplyAmountx USSS ðSupplyAmountx Þ ¼ 100  MaxSupplyAmount

ð9Þ

MSS is determined by the profit margin. For every manufacturer, profit margin is a vital measurement of benefit. Gaining the maximum profit margin is the business goal for factories. Hence, the MSS utility function is:

UMSS ¼ 100  SalesCost ; and FSALES ðProductAmountPriceÞ Sales ¼ ProductAmount  Price

ð10Þ

It is assumed that there is no preference bias amongst each individual in the same stream. Therefore, all resource allocations are equally competitive. Allocation* is defined as the fairness principle allocation. Compared with one resource allocation to the others, the Allocation’ with a higher satisfaction of the four schemes than any other Allocationx can be found as the Pareto-optimum. The formula of the total utility which is taken from a global view which is from the perspective of the whole supply mesh is shown below:

Inf Syst Front

UtilityðAllocationx Þ ¼ UGSS ðAllocationx Þ ^ UCSS ðAllocationx Þ ^ UMSS ðAllocationx Þ ^ USSS ðAllocationx Þ

ð11Þ

M is the marginal minimum acceptable utility. For each utility function, M is set as a different value by the individuals, thus Pareto-optimal allocation shall meet the conditions:   UtilityAllocation*   UtilityðAllocationx Þ Utility Allocation* ¼ MaxðUtilityðAllocationx ÞÞ UCSS ðAllocationx Þ  MCS

objective can be defined as: 1) having the shortest waiting time for the customers’ and suppliers’ orders to optimize manufacturing; 2) delivering a good profit margin and on-time arrivals and 3) producing the lowest total cost and ensuring Pareto satisfaction. The centralized agent, JIT-CA, is responsible for minimizing the waiting time by incorporating the JIT mechanism. The PAs maximize utilities and minimize costs by allocating jobs to different participants in various ways. Ideally, both types of agents collaborate to find the best group of participants with which to form a supply chain as a way of assigning jobs.

By using the Pareto-optimal principle, the resource allocation result is preferred by all participants, so that no one will be defeated in the competition consequently. That is to say, the Pareto-optimal principle plays a significant role in seeking a win-win situation in the supply chain sourcing. 3.6 Fairness and utility measurement Fairness of supply chain resource allocation is measured by the percentage of entities that are able to gain a job. In DC-ASCE resources are allocated according to the Cost-driven principle, as a result only the entities with the lowest cost can gain a job in the supply chain. In CC-ASCE the allocation is on the basis of the Pareto-optimal principle so that everyone is able to gain a job. Fairness ¼

GainJobEntityNumber TotalEntityNumber

ð12Þ

Utility is the positiveness of resource allocation, which represents satisfaction of the supply mesh. UTotal ¼

j X

i i i i UGSS þ UCSS þ UMSS þ USSS

ð13Þ

i¼1

where j is the number of supply chain in resource allocation result. The aim of the optimization is to maximize the total utility of the whole supply mesh (that should have comprised of every participant’s satisfaction in different extends), and to minimize the total cost of the whole supply mesh. Extreme results such as highest utility (satisfaction) or lowest cost may not be achieved in optimal situation; but equilibrium would be established. The CSET model aims to balance the utility and welfare of all of the participants in a supply mesh to establish an optimal supply chain formation. The utility is reflected by the due-date, amount and profit for different types of participants. Welfare is represented by the cost of forming a supply chain. The following pseudo code shows the working logics of the CSET model that embraces the above conditions. The following algorithm is trying to choose a suitable supply chain formation in the supply mesh. The term “suitable” in this

4 Experiment–comparison of CSET and others 4.1 CSET model versus Non-Pareto The Pareto-optimal algorithm comes from theories that originated in the field of welfare economics. Its goal is to search for a resource allocating solution in a supply mesh that allows every participant to avoid suffering a loss. This absence of loss means that every participant, at the very least, achieves their minimum level of satisfaction, which is defined as a mix of costs and other factors being met. Therefore, every participant in the entire supply mesh benefits from the CSET model’s solution. This win-win situation is the result of the supply chain resource allocation. The non-Pareto allocating principle uses a cost-driven method to find the solution with the minimum cost for a particular individual supply chain as the result. The experiments here target to verify the characteristics of CSET that is designed to obtain a win-win situation, while non-Pareto model aims at achieving the lowest price.

Inf Syst Front

4.1.1 Distribution comparison The results generated by our simulation experiments produce a visualization of the relationship between the total cost and participants’ preference as shown in Fig. 6. The utility function is calculated by (11). This relationship can be viewed or clustered into two groups. The first zone obviously has a lower total cost, but its utility (which is sometimes interpreted as satisfaction) is low as well. A worse scenario, such as the one illustrated in the first part of the graph in Fig. 6, reveals utility values that are close to zero, which means that some of the participants have no jobs. These lower total cost allocations are only formed by two single supply chains, while those of higher total costs are formed by three single supply chains. In this case, SC2 is being deprived. Figure 7 shows a comparison of the relevant allocation cost, namely the total cost and average cost of each single supply chain, respectively. The total cost is the sum of costs of all single supply chains in (1)–(5). The average cost is the average cost of each single supply chain. Figure 8 illustrates a clearer evolving trend of average cost which is the total cost divided by the number of participants. Before around the 420th allocation, the formation of only two single supply chains is being considered. This is the non-Pareto zone where some of the participants suffer a lack of jobs. After that point, the utility change corresponds to the zone, as shown earlier in the latter part of the graph in Fig. 6 where all participants have gained jobs. The nonPareto average cost has three distinct zones that correspond to Sets A, B and C in Fig. 9. Figure 10 is MathLab’s visualization of the results generated by the simulation experiment conducted and reported in the previous section. For each single supply chain there is also a frontier that specifies the utility values. For example, for supply chain one (SC1) the total utility experiences a dramatic change at approximately the 420th possibility. Before reaching this point, the utility value constantly changes from 0 to about 300. After this point, the change

Fig. 6 Visualization of the third supply chain

Fig. 7 Total cost comparison

becomes relatively more stable at around 300. It is interesting to note that this point also exists along the other two single supply chains, SC2 and SC3. Figure 9 shows why this point exists. Among the 1,296 possibilities of resource allocation, the results can be clustered into four sets in relation to the utilities of the single supply chains, each of which has three-dimensional values, specifically the utility of every single supply chain (SC1, SC2 or SC3). Rotating the view of the 3D model reveals that Set B is distinct from the other three sets, namely sets A, C and D. The values for all dimensions of Set B are believed to be non-zero with the exception that at least some of the values in one of the three dimensions are believed to be zero. For this reason, the first zone in Fig. 10 is composed of the elements in sets A, C and D, which are not preferred by all of the individuals in the supply mesh. The visualization results reveal which Pareto zone is favored by all the participants. The procedure for finding the Pareto optimum is shown in Fig. 11. The second zone, which is near the lower end of the chart in Fig. 6, has a much

Fig. 8 Average cost comparison

Inf Syst Front

Fig. 9 CSET model result distribution

higher total cost because everyone received a preferable job from the resource allocation. Although the total cost is much higher in this zone, the average cost is not at its highest compared to the non-Pareto zone presented in Fig. 7, which falls into the mid-range of all average costs shown in Fig. 8. Meanwhile, every local utility of the three single supply chains is not zero, as shown in Fig. 10, which means that the individual satisfaction must not be zero. Set B from Fig. 9 represents the non-zero utility elements. If the set of Pareto-optimum results is defined as Set P, then Set P ⊂ Set B. As the simulation results show, 96 of the 1,296 successful resource allocations meet the utilities requirements of the local schemes. The distribution of the 96 successful

allocations is shown in Fig. 12. It can be seen that there are only 21 points in this chart because many of the results among the 96 allocations actually overlapped. A frontier of maximum total local scheme utility exists in the 3D distribution. Its value is 938 (308+309+3210938). After bunching the duplicated results, there are only 6 unique resource allocations, which give a net success rate of 0.463 % of 1,296 possible allocations in this experiment. In other words, every participant can gain a preferable job with maximum satisfaction on this plane. If this place is defined as Set S, then Set P ⊂ Set and S ⊂ Set B. In addition, Paretooptimum allocations exist on this plane and are defined as the points with minimum total cost and maximum global satisfaction.

Inf Syst Front

Fig. 10 The total utility and local utility of the three single supply chains

4.1.2 Value comparison As a further analysis, we use the following tables and charts to show the differences between non-Pareto distribution and CSET Pareto-optimum distribution. As Fig. 13 shows, we can know that the Pareto-optimum result has a relatively high manufacturing cost. Fortunately, this high cost is still below the maximum non-Pareto cost. This high manufacturing cost exists because all the factories have jobs to do, which inevitably increases the total manufacturing cost. In Fig. 14, the Pareto-optimum result has a fixed cost that equals the maximum of non-Pareto. Similar to the cause of high manufacturing costs, namely fulfilling every client’s order, the factories have to procure enough raw materials.

Fig. 11 Flow chart of Pareto-seeking

For this reason, the suppliers will provide all the raw materials they can so that the fixed cost is improved until all the supplied resources are used up. Consequently, all of the suppliers provide raw materials for the supply chain so that no suppliers will be deprived. In Fig. 15, the high cost of the Pareto-optimum result is similar to that for the non-Pareto result. To satisfy every individual in the supply mesh, the total product amount and total raw materials amount are at high required quantities. Moreover, the deliveries occur among all participants. Hence, the total freight cost of the Pareto optimum is large. In terms of JIT-Pareto integration, Fig. 16 shows that the penalty cost is kept at a consistently low level. Figure 17 shows that Pareto optimum has a large total cost. In particular, some may challenge the feasibility of Pareto-optimum allocation. Generally, the reward of such high total cost comes with the advantages in sales and individual preferences. Figure 18 shows that using up all the resources to ensure that everyone is able to get a job with a high level of satisfaction results in the maximization of total sales of Pareto-optimum allocation. For a factory, these larger sales mean that its profit rises up, assuming the cost remains the same. Therefore, this situation is naturally favored by manufacturers. In Fig. 19, the Pareto-optimum allocation not only gains much better utilities, which are preferred by everyone, but it

Inf Syst Front Fig. 12 Distribution of suitable allocation by local utilities

also clearly obtains an optimized resource allocation with maximum happiness for all participants. 4.2 CSET model versus Non-Pareto Pareto originally used the means of even distribution to manipulate the allocation of wealth among individuals, because it seemed to effectively illustrate how a larger portion of the wealth in any society is owned by a smaller percentage of the people in that society (Koch 2001). The Pareto

Fig. 13 Non-Pareto vs. Pareto: manufacturing cost calculated by (2)

principle was often expressed by the phrase, “20 % of the population controls 80 % of the wealth.” However, this principle is somewhat different from our proposed Paretooptimality in the CSET model. Figure 20 shows that among the 12 experimental participants in a supply mesh with the 80/20 Pareto allocation result, only 3 out of the 12 (25 %) participants had high utilities above 80 %. In other words, only 25 % of the participants obtained more than 80 % utility, corresponding

Fig. 14 Non-Pareto vs. Pareto: fixed cost calculated by (1)

Inf Syst Front

Fig. 17 Non-Pareto vs. Pareto: total cost Fig. 15 Non-Pareto vs. Pareto: 3PL freight cost calculated by (5)

to the 80/20 Pareto principle. The result of this 80/20 principle is that only 5 out of 12 (41.67 %) have utilities at more than 60 %, which means less than half of the participants are able to gain utilities of 60 % or more. In addition, some of the participants’ utilities are as low as zero. Obviously, this result is unfair because the majority of the utilities are absorbed by a minority of the participants. Although the 80/20 Pareto principle is considered common sense in business management, it will cause destructive competition in a supply chain business. Compared to the 80/ 20 Pareto allocation result, the contribution of our proposed model is to eliminate this kind of unfairness. Our model attempts to ensure that all individuals’ utilities are improved through the implementation of the CSET model. In our simulation of CSET Pareto, 5 out of 12 participants (41.67 %) have utilities of more than 80 %. Simultaneously, all participants have their utilities increased by approximately 60 %. 4.3 Co-operative versus destructive competition

the Pareto optimum. The cost is defined as the total operating costs for all the operational supply chains. To some extent, people may question the practicability of the proposed model. However, every coin has two sides: the high cost in the CSET model is compensated by the best satisfaction and fairness. Figure 21 presents the simulation result of a three-tier supply mesh under environments of both Destructive Competition and Cooperative Competition. There are five distributors, five factories, five raw material suppliers and one 3PL. A Java-based analysis tool is developed. The supply chain formation used as input to the simulator is derived from (Nowell 2005), which closely resembles data of a realistic supply chain in a Global Textile Supply Chain in the US. The top of the visualization shows the supply chain resource allocation of Cost-driven principle. It shows the cost and satisfaction on every entity label. For the cost, on the distributor’s, cost is the delay cost of penalty and inventory that the factory has to pay for; on the factory’s, cost is the manufacturing and all freight cost that the factory has to shoulder itself; and on the supplier’s, cost reflects the fixed cost of raw material procuring. For the relevant calculation

The comparative analysis above shows that the cost of normal allocation (non-Pareto) is much lower than that of

Fig. 16 Non-Pareto vs. Pareto: penalty cost calculated by (4)

Fig. 18 Non-Pareto vs. Pareto: sales

Inf Syst Front

Fig. 19 Non-Pareto vs. CSET: utility comparison calculated by (11)

refer to Section 3.4. A. Satisfaction baseline is also presented on the labels. As a result, only one entity in each stream is able to get a job (Distributor 5–Factory 1–Supplier 2), with the minimum total cost of $8,946. The total utility of the three participants is 39. However, in the light of fairness and total utility calculation formulas mentioned in Section 3.7, only 20 % (3 of 15) entities get jobs. Compared with Cost-driven principle allocation, the lower visualization shows that the percentage of job gaining under Pareto-optimal principle is as high as 100. Five sub-

Fig. 20 80/20 Pareto principle vs. CSET model

supply chains are formed consequently. Although the cost of each sub-supply chain is slightly higher than that of the Cost-driven result, total utility is greatly improved from 39 to 189. In addition, two bar charts (both cost and satisfaction comparisons) are displayed on the right. To further compare these two kinds of allocations, we also consider the long-term running when the number of entities in each tier increases (from 2 to 7) in Tables 1 and 2. Pareto-optimal principle brings an absolute fair outcome where everyone is able to gain a job all the time. However, the Cost-driven principle only selects the lowest cost for allocation. Hence, as the number of entities increases, the percentage of jobs gained decreases in DC-ASCE (in Fig. 22). Figure 23 shows the utility comparison. When the number of entities grows from 2 to 7, the utility of Cost-driven principle is at a stable but very low level (approx. 200). Obviously, the utility of the Pareto-optimal principle keeps a stable growing trend along the polynomial trend-line. This Pareto-optimal situation remains even if the number of entities increases continually. “Survival of the fittest” is one of the natural rules of life according to Darwin’s theory of evolution, which has been applied to various economical business models. Nevertheless, individuals have proven that collaborative power can harness aggressive competition and evolve into mutually beneficial exchange. Many have observed that when things are desired in common by many individuals, this desire generally includes mutual survival, which encourages cooperation rather than conflict (Davidson and Rees-Mogg 1999). This interest in mutual survival is of vital importance to win-win consequences in supply chain businesses. In our opinion, the experimental results reflect two extreme business phenomena: destructive competition and cooperative competition (Smith 1950). The former type of competition is aggressive by nature. A competitive company, for example, will try to monopolize all the jobs by offering low prices. The latter type is cooperative and often results in a win-win situation. The non-Pareto process is classified under destructive competition. On the one hand, its cost is much lower and some of the weak opponents are defeated. Only a minority of the participants can gain with the majority of the utilities, so it is an unfair resource allocation scheme for the majority of participants. On the other hand, non-Pareto allocation is a cost-driven solution that eliminates high-cost participants. To this end, this allocation is not preferred by those individuals who cannot obtain jobs. It opposes the desire for mutual survival and represents a “winner takes all” situation. This rationale, which is being challenged, is a zero-sum game: the success of one group is dependent on the failure of the other competing groups. The consequence is a binary winner-loser game. For those who fail to obtain jobs, this

Inf Syst Front

Fig. 21 Visualization GUI of CSET simulator with entity number 0 5

allocation is unwilling. A non-Pareto allocation scheme may be suitable for short-term/one-time supply chains without any fundamental intention of cooperation in an eMarketplace. Those seeking a long-term profit, however, especially if a supply mesh that is meant to stay in a market

for any length of time, will find that this type of competition results in many companies suffering losses as their survival is threatened. Unlike destructive competition, CSET Pareto-optimality coordinates the competition so that everyone gains from the

Table 1 Parameter inputs Supply mesh Client (retailer)

Manufacturer

Supplier

Logistic provider

Final product amount

Final Product Amount

Unit Delivery Time

Anticipant delivery time Latest acceptable delivery time Location Daily penalty Final product price

Raw Material Amount Manuf. Ratio Location Daily Manuf. Cost Daily Inventory Cost Working Duration

Max Raw Material Supply Amount Supply Time Raw Material Price Location

Unit delivery Price

Inf Syst Front Table 2 Simulation results for long-term operation of supply mesh under two extreme schemes

Entity#

2

3

4

5

6

7

16520 219 50 16840 215 20218 204 – – –

15600 200 33 17860 204 17270 192 17440 195 –

16070 194 25 20350 191 17910 196 18780 184 20020

15870 191 20 20900 174 17080 185 19780 198 16370

16180 198 16 19490 194 19780 182 17280 190 15830

16080 197 14 20230 198 19860 193 20850 194 16820

– – – – – – – 100

– – – – – – – 100

203 – – – – – – 100

201 18290 187 – – – – 100

192 22710 191 18100 183 – – 100

186 17050 206 16540 179 19590 203 100

Attribute DC

CC

Cost Satisfaction Job% SC1 Cost Satisfaction SC2 Cost Satisfaction SC3 Cost Satisfaction SC4 Cost SC5 SC6 SC7

Satisfaction Cost Satisfaction Cost Satisfaction Cost Satisfaction

Job%

recourse allocation. In this case, the co-operative competition exists within a supply chain. Everyone benefits to survive in a competitive environment. The expense of obtaining this upstanding condition is a relatively high total cost with an acceptable ceiling, which means that individuals are willing to trade the disadvantage of this high cost for mutual survival. For instance, in our experiment the manufacturers were willing to bear a relatively high cost on the basis of a no less than 50 % profit margin (average resulted is 58.67 %). Although some allocations can deliver the factory’s maximum profit margin as much as 68 % in nonPareto, the profit margins of others in the same allocations is zero because there is no job distribution. Likewise, 73.67 %

and 79.67 % satisfaction for the fabric suppliers and yam suppliers, respectively, are included in the Paretooptimum result, even though some of the individuals give up the desire for a higher level of satisfaction. Thus, the proposed CSET model should be more suitable for long-term supply chains in co-operative competition environments. In summary, the results of the experiment shows two types of competitions in an e-marketplace supply chain, and Pareto-optimum resource allocation plays a significant role in achieving a win-win situation.

CC-ASCE

DC-ASCE

Poly. (CC-ASCE)

1600 1359

1400 1132

Satisfaction

1200 945

1000 774

800 591 600 420 400 219

200

194

191

198

197

200 0 2

3

4

5

6

Entity Number in each Stream

Fig. 22 Jobs-gaining percentage comparison in a long-run

Fig. 23 Total utility comparison in long-term

7

Inf Syst Front

5 Conclusion

References

Our previous work proposed an optimized supply chain framework called CSET (Yang et al. 2010). This paper proposes the workflow and processes in details that CSET model can essentially provide mutual benefits to companies by collaboratively forming dynamic supply chains online. CSET works by utilizing an intelligent agent system in a membership-based eMarketplace consortium known as a supply mesh. The most innovative feature of the proposed CSET model is the integration of the Pareto-optimal algorithm and the JIT principle into an intelligent agent system. Its ability to find Pareto-optimal job allocations means that the resource allocation is preferred by all participants, leading to win-win solutions resulting from the formation of the right supply chains in the supply mesh. Incorporating the JIT principle means that the waiting and information transferring times are significantly shortened. This facilitates the faster and more effective flow of supply chains via pipelining. Furthermore, every company in a supply chain is able to get a job through the proper coordination exercised by the CSET agents. When viewed from a supply chain evolution perspective, the CSET model is applicable for evolving existing e-Marketplaces by equipping them with features that promote the formation of dynamic supply chains and ensure the common welfare of the participants. The optimization of resource allocation insures the costs and growth for both the individual companies in the supply mesh and the supply mesh as a community. No participants will suffer losses, thanks to the Paretooptimal algorithm in CSET. To verify the feasibility of the CSET model, a Java Bean + Java Server Page (JSP) simulator was developed to simulate the supply chain resource allocation as a case study. By simulating different supply chain formation modes, be they cost-driven or Pareto (cooperative or otherwise), we can see that the CSET model does yield better resource allocations that are evidenced by the resultant costs and utilities. The output result shows that the supply chain formation is the most satisfactory for all participants at an acceptable total cost. Under the destructive competition scenario, the lowest cost resulted from the cost of sacrificing many weak companies. In a co-operative competition scheme our results show that the CSET model participants have to spend a slightly higher total cost than those under destructive competition, but as long as all of the participants survive, this proves a win-win situation in the long-term. The CSET model is definitively able to obtain the desired balance between welfare and utility.

Banker, S. (2005). The global, make-to-order supply chain: Is it time to examine alternative models, ARC insights. Dedham: Allied Drive. Billesbach, T. J. (1991). A study of implementation of just-in-time in the United States. Production and Inventory Management Journal, 32(3), 1–4. Buyya, R., Abramson, D., Giddy, J. (2000). An economy driven resource management architecture for global computational power grids. Proceeding of International Conference on Parallel and Distributed Processing Techniques and Applications. Chavez, A., & Maes, P. (1996). Kasbah. An agent marketplace for buying and selling goods. Proceedings of the First International Conference on the Practical Application of Intelligent Agents and Multi-Agent Technology. Cook, R. L., & Rogowski, R. A. (1996). Applying JIT principles to process manufacturing supply chains. Production and Inventory Management, 1st Quarter, 12–17. Davidson, J. D., & Rees-Mogg, W. (1999). The sovereign individual: Mastering the transition to the information age, 380. Fudenberg, D., & Tirole, J. (1983). Game theory, MIT Press, chapter 1, section 2.4. Harland, C. M. (1996). Supply chain management: Relationships, chains and networks. British Journal of Management, 7, 63–80. Hobbs, O. K. (1994). Application of JIT techniques in a discrete batch job shop. Production and Inventory Management, 1st quarter, 43–47. Holsapple, C., Lai, H., & Whinston, A. (1995). Analysis of negotiation support systems: Roots, progress, and needs. The Journal of Computer Information Systems, 35(3), 2–11. Holsapple, C., Lai, H., & Whinston, A. (1997). Implications of negotiation theory for research and development of negotiation support systems. Group Decision and Negotiation, 6(3), 255–274. Holweg, M., & Pil, F. (2004). The second century: Reconnecting customer and value chain through build-to-order. Cambridge, MA and London, UK: The MIT Press. Homburg, C., & Schneeweiss, C. (2000). Negotiations within supply chains. Computational & Mathematical Organization Theory, 6 (1), 47–59. Homburg, C., Koschate, N., & Hoyer, W. D. (2006). The role of cognition and affect in the formation of customer satisfaction: A dynamic perspective. Journal of Marketing, 70(3), 21–31. Kim, H. S., Cho, J. H., Choi, H. R., Hong, S., Kang, M. H. (2006). Optimal supply chain formation using agent negotiation in a SET model-based make-to-order. Proceedings of International Conference on Electronic Commerce, 579–583. Koch, R. (2001). The 80/20 principle: The secret of achieving more with less. London: Nicholas Brealey Publishing. Kropotkin, P. (2005). To alter or to abolish, chapter 5: All against all, Darrell Anderson. Lau, H. C., & Zhang, L. (2004). A two-level framework for coalition formation via optimization and agent negotiation. Proceedings of the intelligent Agent Technology, IEEE Computer Society, 441–445. Lau, H. C., Zhang, L., Liu, C. (2005). Solving generalized open constraint optimization problem using two-level multi-agent framework. Proceedings of the IEEE/WIC/ACM international Conference on intelligent Agent Technology (September 19–22, 2005). IEEE Computer Society, 558–564. Li, C., & Li, L. (2012). Collaboration among mobile agents for efficient energy allocation in mobile grid. Information Systems Frontiers, 14(3), 711–723. Li, H., Ahn, D., Hung, P. C. K. (2004). Algorithms for automated negotiations and their applications in information privacy. Proceedings of the IEEE International Conference on E-Commerce Technology, 255–262.

Inf Syst Front Lim, L., & Benbasat, I. (1993). A theoretical perspective of negotiation support systems. Journal of Management Information Systems, 9 (3), 27–44. Liu, R., & Kumar, A. (2011). Leveraging information sharing to configure supply chains. Information Systems Frontiers, 13(1), 139–151. Lo, G., & Kersten, G. E. (1999). Negotiation in electronic commerce: Integrating negotiation support and software agent technologies. Proceedings of 5th Annual Canadian Operational Research Society Conference. Monden, Y. (1981). Adaptive Kanban system helps Toyota maintain just-in-time production. Industrial Engineering, 13(5), 29–46. Nam, I. H. (2003). Benefit of supply chain coordination. Seoul Journal of Business, 9(1). Nowell, C. H. (2005). Market competitiveness in the global textile supply chain: Examination of supply chain configurations. MSc Thesis, North Carolina State University. Putten, S., Robu, V., Poutré, H., Jorritsma, A., Gal, M. (2006). Automating supply chain negotiations using autonomous agents: A case study in transportation logistics. Proceeding of International Conference in Autonomous Agents, the International Workshop on Agent Theories, 1506–1513. Rangaswamy, A., & Shell, G. R. (1997). Using computers to realize joint gains in negotiations: Toward an electronic bargaining table. Management Science, 43(8), 1147–1163. Smith, A. (1950). An inquiry into the nature and causes of the wealth of nations. London: Methuen. Temponi, C., & Pandya, S. Y. (1995). Implementation of two JIT elements in small-sized manufacturing firms. Production and Inventory Management Journal, 3rd Quarter, 23–29. Tian, J., Foley, R., Yao, X., Tianfield, H. (2006). An extended contract net mechanism for dynamic supply chain formation and its application in China petroleum supply chain management. Multiagent and Grid Systems, 183–207. Walker, W. (2004). “Supply chain flexibility”. Montgomery Research ASCET 6. Wigand, R. T., & Benjamin, R. I. (1995). Electronic commerce: Effects on electronic markets. Journal of Computer-Mediated ComMunication, 1(3).

Yang, H., Fong, S., Zhuang, Y. (2008). Applying Pareto-optimal and JIT techniques for supply chains. Proceeding of International Conference on Hybrid Intelligent Systems, 308–313. Yang, H., Fong, S., & Zhuang, Y. (2010). CSET automated negotiation model for optimal supply chain formation. World Review of Science, Technology and Sustainable Development (WRSTSD), 7(1/2), 67–78. You, F., & Grossmann, I. E. (2007). Optimal design and operational planning of responsive process supply chains. Process Systems Engineering, Wiley.

Hang Yang He is a PhD candidate at the University of Macau. He obtained an MSc (First Honor) in Electronic Commerce Technology from the University of Macau in 2009; and a Bachelor’s degree in Economics and Electronic Commerce from Guangdong University of Foreign Studies (China) in 2007. He worked for the companies of Fortis Insurance and China Petrol in Hong Kong and Beijing. Dr. Simon Fong supervises him and his research interests cover data mining, business intelligence, electronic commerce, and web intelligence. Simon Fong He graduated from La Trobe University in Australia, with a First Class Honours BEng, a Computer Systems degree, and a PhD. Computer Science degree in 1993 and 1998 respectively. Simon is now working as an Assistant Professor in the Computer and Information Science Department at the University of Macau. He is also one of the founding members of the Data Analytics and Collaborative Computing Research Group in the Faculty of Science and Technology. Prior to joining the University of Macau, he worked as an Assistant Professor in the School of Computer Engineering at Nanyang Technological University in Singapore. Before his academic career, Simon took up various managerial and engineering posts, such as being a systems engineer, IT consultant, integrated network specialist, and e-commerce director in Melbourne, Hong Kong, and Singapore. Some companies that he worked at before include Hong Kong Telecom, Singapore Network Services, AES Pro-Data, and the United Overseas Bank in Singapore. Dr. Fong has published over 150 peer-reviewed international conference and journal papers, mostly in the area of e-commerce technology, business intelligence, and datamining.