Multilateral Negotiations in Distributed, Multi-agent ...

3 downloads 0 Views 275KB Size Report
the multilateral negotiations are suitable for exchange of commodities that are ... platform for multi-commodity exchange, and (iii) evaluation of proposed inter-.
Multilateral Negotiations in Distributed, Multi-agent Environment

Piotr Paªka

Warsaw University of Technology, Nowowiejska 15/19, 00-665 Warsaw, Poland [email protected]

The paper describes the problem of multilateral negotiations, conducted in a distributed environment. In the multilateral negotiation, more than two partners participate simultaneously. Multilateral negotiation processes are important in the case where objects of the negotiation are bounded with by certain constraints (e.g. physical constraints), or if the participants trade with dierent bundles of commodities. The commodities structure may then force the necessity of multilateral negotiation. The interaction protocol for the multilateral negotiation, based on FIPA's protocol, is presented. A simple, yet representative example is presented to illustrate the problem. The results of computer simulations of the proposed negotiation process are presented. Abstract.

multilateral negotiation, distributed trade, communication protocol, multi-agent system, multi-commodity trade Keywords:

1

Introduction

In e-commerce complex markets (e.g. infrastructure markets, where exchange of goods takes place in a complex network structure), there are numerous limitations and other ties between actors  market participants, and between objects of trade  commodities. Frequently occurring constraints arising from a structure of the problem, as well as preferences of participants, cause inability to reach a satisfactory outcome in bilateral negotiations. The solution is either the trade on centralized auctions or exchanges (is there exist one) or the multilaterally negotiated agreement. For certain markets (e.g. for the energy market) the centralized trade exists, where the transmission system operator (TSO), which manages the infrastructure, ensure fulllment of the associated constraints, and is also the organizer of the trade, exists. However, the predetermined market structure, which allows for the centralized exchange of commodities, often does not exist. Thus, the multilateral negotiations are suitable for exchange of commodities that are not covered by auctions or exchanges. To implement the negotiation process, the proper interaction protocol should exist. In the literature we can nd the simple protocol for multilateral negotiation [3], which is generalization of the protocol for bilateral negotiation [15]. We propose another implementation of the interaction protocol for the multilateral negotiations. We do not assume the simplicity of the proposed protocol,

2 but its eectiveness, correctness and usefulness. The protocol is consistent with FIPA organization specications [4]. To verify the characteristics presented, we implement the negotiation process in the multi-agent trading platform [7,11], and evaluate it on the simple example of the multilateral trade. Note however, that the simplicity of the example does not preclude the wide application of the protocol in real negotiations. The contributions of this paper are threefold: (i) development of the interaction protocol for multilateral negotiations, consistent with FIPA's specication (ii), implementation of the multilateral negotiation protocol in the multi-agent platform for multi-commodity exchange, and (iii) evaluation of proposed interaction protocol for multilateral negotiations.

2

Multilateral Negotiations

Participants on the distributed market are directly engaged in exchange of goods, by negotiating the best, from their point of view, contracts. Various participants reach bilateral or multilateral agreements on such markets, usually after complex negotiations. The simplest type of agreement is a bilateral agreement. Note however that in the case of multi-commodity trade (e.g. simultaneous energy and ancillary services trade) the agreements may aect not only many goods, but also a number of suppliers and customers. Multilateral negotiations are most common in situations where interests of various social groups, companies, organizations, political parties or even states converge, and a resource or resources, which are involved in negotiations, are the common good. An example of the multilateral negotiation concerns establishing limits on greenhouse gases emissions. During the Kyoto conference in 1997 representatives of countries agreed on the need to react on the global warming [10]. In order to ensure the fulllment of the global constraints, each country declared the level of the CO2 emission. The problem of water allocation among farmers in Adour basin in southwestern France is described in the work [16]. Since water uptake in the upper reaches of the river aects the ow in its lower course, the negotiation process, which consists in assigning participants to dierent values of water consumption, and charging them fees, is a process that requires the activity of all participants. In [1] the model for multilateral negotiation, on the example of negotiations between the three entities, is presented. The negotiation concerns the location of a distribution center. Authors apply the non-cooperative multilateral bargaining model developed in [14]. Multilateral negotiations may be also enforced by a simultaneous trade in more than one commodity, so called multi-commodity trade [5]. The multi-commodity trade facilitates compliance with complex preferences of market participants. In the paper [12] authors discuss the problem of multilateral negotiation in an environment with the centralized architecture. The paper considers dierent mechanisms for communication, however related issues are similar. Author of [3] considers the simple monotonic concession protocol for multilateral negotiation. Author generalizes the bilateral

3 monotonic concession protocol. The monotonic concession protocol was developed in the work [15], it is a generalized version of the Zeuthen Strategy [19]. The same protocol was applied in the work [17], where authors consider the negotiation on the production sequencing in the car factory.

3

Case study

To illustrate the problem of the multi-commodity trade that requires the multilateral negotiation, we use a fairly simple example. Let us assume the existence of three building plots on the real estate market (see Fig. 1). Entity B is the owner of building plot X, entity C is the owner of parcels Y and Z. Entity B wants to exchange parcel X to Z, because it is consistent with his construction plans. He allows the possibility for selling property X, and purchasing plot Z, but entity C will not agree to sell the Z building plot separately, because he knows that parcel Y is too small. Thus, entity C wants to sell bundle of parcels Y and Z. Entity B does not have enough money to buy parcels Y and Z, besides parcel Y is useless for him. Entity A wants to buy the bundle of X and Y plots, because he is interested in building a house on plot X, and a farm building on parcel Y.

Plan of building plots, which are subject to negotiations. X, Y, and Z are the names of of individual parcels Fig. 1.

Note, that no pair of entities can achieve acceptable to both parties agreement. However, the agreement between three entities, meets the preferences of all of entities. Thus, in this case the multilateral negotiations are justied by the preferences of all participants.

4

Interaction Protocol for the Multilateral Negotiations

In this section we present the protocol for the multilateral negotiation. Trading processes on dierent markets, related to negotiation or bidding, are specic for each type of trade. Multi-agent systems [18] are natural framework for automating tasks of the trading processes in the environment, where a number of entities, guided by its own interests, exists, and identify themselves with their own objectives [7]. In the multi-agent system, an agent is a software component which is able to take autonomous actions and conduct negotiations, by submitting offers and exercising sophisticated strategies. However, trading processes can be performed in various circumstances and may take dierent forms.

4

4.1 Agents As we assume that the negotiation is performed using the multi-agent system which models a distributed environment, the information is distributed between particular agents who participate in a trade. The roles structure of the agents considered in the project of multi-agent platform for the multi-commodity exchange [11] is wide. Since the multi-agent platform is a framework, sometimes it appears that the roles structure of the agents should be expanded or claried. The individual roles are usually closely tied to the application, thus we describe only those roles, which are needed.

Negotiator agent

The negotiator agent realizes the best contracts from its prefer-

ences point of view. Agreement is preceded by (possibly complex) negotiations. Negotiator agent has wide freedom in applying negotiation strategies. It should be able to negotiate multilaterally. A particular negotiator agent, to be able to negotiate, needs either to inform other agents about its negotiation proposals or collect a specic information about other negotiators, their proposals, and particularly about the possibilities for negotiation. We can certainly imagine a case in which each agent broadcasts message about a negotiation proposal to all remaining agents. However, in large systems, such a solution would involve a massive communication eort. Moreover, agents would obtain a huge amount of unnecessary information.

Morris Column agent

In the centralized system it is possible to create a central

repository, which collects all the necessary data about the agents, and existing and planned trade processes. As distributed systems do not have (and often cannot have) any central entity, such repository cannot be built, so there is a need to share the information in a dierent way. For this purpose we dene a special agent called Morris Column agent [5]. The task of this agent is to oer some public location, where other agents may hang a notice to report certain information about provided processes of trade. Additionally the Morris Column agent should provide a functionality of searching, ltering and removing the information. In a broader context, agents may leave notices on initiating various types of negotiation processes. Despite apparent compliance with the Morris Column agent, Directory Facilitator agent (provided by FIPA specication) operates in limited way. It accepts queries as a simple strings, interpreted by it as the names of the services registered by other agents, and returns only identiers of matched agents. The Morris Column agent accepts complex queries, mostly formulated in XPath language (other languages are also allowed, e.g. XQuery), and returns XML formated response. The Morris Column agent provides information not only about agents, but also about other aspects of the trade.

Broker agent

After the negotiator agents submit to the Morris Column their

negotiation proposals, there is a need to associate them. The broker agent associates the matching negotiation proposals, and then noties the agents (which submitted the proposals) about the possibility of agreement. Broker agent has

5 certain algorithms for matching proposals. One of it is presented in further parts of the paper. In this paper we assume that the broker agent has limited functionality, nevertheless we do not limit its functionalities to only those activities. In [2] authors consider the negotiation host agent, which has very similar goals as the broker agent. Nevertheless, the negotiation host agent is a central entity in the negotiation process. It creates new negotiator agents and collects proposals. In our proposition, these operations are performed by other agents. Also we assume that more than one broker can exist. Application of these three agent roles makes it possible to perform multilateral negotiations.

4.2 Communication in Distributed Environment The Foundation for Intelligent Physical Agents (abbrev. FIPA) [4] developed the interaction protocol for the bilateral negotiations  the

Interaction Protocol

Iterated Contract Net

[2]. Unfortunately, this interaction protocol is insucient

for performing the multilateral negotiation. The new interaction protocol for the multilateral negotiation was developed. We based it on the modication of the

Iterated Contract Net Interaction Protocol,

described in [13]. The modication

relates to allow the bilateral negotiations in distributed environments. Particular stages of the developed interaction protocol for multilateral negotiations are presented.

Submission of proposals A number of negotiating agents submits the negotiation proposals (the call for proposal message) to the Morris Column agent (see Fig. 2). We assume the application of the Morris Column agent instead of the Directory Facilitator agent due to the potential complexity of the proposals, which may contain information about many commodities, prices and possible quantities of commodities. Negotiation proposals are written using the XML-based dialect of

3

M model  M3-XML language [11].

Association of proposals

The broker agent browses familiar Morris Column

query-ref message  see Fig. 2). inform message), it tries to associate them

agents, for the new negotiation proposals (the After each new proposal received (the

in such a way that agents could potentially reach an agreement. The proper negotiation begins after the number of agents are associated. The simple algorithm for matching oers is presented in section 4.3.

Proper negotiation

A proper negotiation is started by the current negotiation

manager, by sending its proposal to negotiation participants (the

propose

mes-

sage). If the manager is willing to sell commodities, the proposal includes the value which he wants to receive. If the leader is willing to buy commodities, the proposal includes the value which it is willing to pay. As set forth in [6], we assume that the purchase value is negative. Every agent, after receiving the proposal, has to decide whether it accepts it. If the agent accepts the received proposal, it will send an

accept-proposal

6

Preparatory phase of the multilateral negotiation: submission of new proposals and oer matching Fig. 2.

Fig. 3.

Multilateral negotiation

message to the manager. If the agent decides that the proposal does not meet

rejectproposal(current) message to the manager. If the agent is no longer interested in its preferences, and it expects a concession from manager, it will send a

the negotiation (e.g. if it reached more protable agreement in other negotiation

reject-proposal(whole) message to the manager. If the manager receives at least one reject-proposal(whole) message, it will nish the negotiation, by sending the inform(nish negotiation) message to every negotiator. If the leader receives an accept-proposal message from all the agents, process), it will send a

it will proceed to the two-phase commit protocol, which is the nal completion of the agreement. If not, it will inform other agents (

inform(continue negotiation) )

that negotiation should proceed.

Two-phase commit

An adaptation of the two-phase commit protocol (2PC), well

known in distributed OLTP (On-line Transaction Processing) systems, can be used as a method for xing agreements on decentralized markets. It guaranties

7 safe completion of the agreement, ensuring that all the involved parties ultimately accept the agreement. This is especially important in situation where agents are simultaneously involved in the dierent processes of negotiation, and each of them may reach better agreement in any moment. We assume that every negotiation process is disobliging until the two-phase commit. From the beginning of the two-phase commit protocol agents x the commodities as contracted. FIPA standards do not directly dene this protocol, but it can be quite easily dened on the base of existing communication acts. In [9,13] an appropriate communication schema is proposed for bilateral transactions, nevertheless the 2PC protocol can also be used for multilateral ones.

4.3 Oer Matching Algorithm The simple oer matching algorithm is presented. This is an internal algorithm of the broker agent. The algorithm creates empty collection of matrices. Each matrix is indexed by the names (unique identiers) of commodities and participants. If the broker agent nds a new proposal on the Morris Column, the algorithm will add the new matrix to the collection. Also all the matrices, in which there is at least one commodity contained in the newly added oer, are duplicated. The algorithm adds to every newly created or duplicated matrix the information about notier's preferences  for each notier the new column is created, where for each commodity +1 is added if the notier is willing to sell this commodity, or -1 is added if the notier is willing to buy this commodity (see Fig. 4).

Fig. 4.

Oer associating algorithm: accepting new oers

If the algorithm detects that it is possible to make an agreement (i.e. for each commodity, the sum over all participants is equal to zero), the broker agent noties the corresponding agents. The algorithm does not include the price bids, as it treats them as the subject of further negotiations.

8

5

Evaluation

The multi-agent platform is implemented in Java. To obtain implementation of FIPA's communication standards it uses the JADE framework. XML-based di-

3

alect of multi-commodity market data model M (M3-XML) is used to notate the content of exchanged messages. To facilitate the use of XML based dialect, the JAXB framework, which enables translation of XML objects to Java classes, is applied [8]. The platform is described in details in paper [11]. Using the platform we present the implementation of interaction protocol for the multilateral negotiation, implement the negotiation scenario, and perform tests. The simulation of the negotiation process was performed. As the example, we use the case study described in section 3. At the beginning of the negotiation, participant A argues that the plots X and Y are worth for him 1 mln $. In fact, he allows to pay for them up to 1.6 mln. $. Participant B wants to receive 200 thous. $ for the exchange the property X on Z. In fact, he allows to pay 100 thous. $ for this transaction. Finally, the participant C, who claimed that he wants to cash in on land sales Y and Z 2.8 mln. $, allows the sell them at 1.5 mln. $. Note, that the transaction can not be completed, taking into account the initial proposals. Thus, each participant concedes its pricing (at each round) of respectively:

100, 50

and

250

thous. $.

Negotiation nished after 18 iterations. During the negotiation process 116 messages have been exchanged. Every negotiating agent sends 37 messages: one

cfp ), 12 messages with propopropose ), 12 messages informing about the result of negotiation process (inform ), 11 messages rejecting current proposal (reject-proposal ), and one accepting message (accept-proposal ). The Morris Column agent sends one message (inform ) to broker agent, broker agent sends one message to the Morris Column (query-ref ), and three messages (propose ) to each negotiating agent. message with submission to the Morris Column ( sition (

Agents involved in the negotiation have reached an agreement allowing for the nalization of the transaction. As a result of the agreement participants A and B respectively have agreed to pay sold the plots for amount of

1.55

1.5

mln. and

50

thous. $, participant C

mln. $.

To analyze duration of negotiation process, ten simulations were performed on the computer with Intel Core i7-920 (2.67 GHz) processor, and 8 GB RAM, under the Debian 5.0.3 OS. Duration of the whole negotiation process, taking into account the time from publishing the negotiation proposals to successful nishing the negotiation, and the duration of the relevant negotiation (from the rst notication of the proposal) has been measured. The time for single iteration of the negotiation also has been measured (see Table 1). Simulation has shown, that the process for the multilateral negotiation proceeds fast, since the execution time of single iteration of negotiations associated with the formation of messages, their transmission, analysis and coordination are the order of fractions of a second. In normal commercial practice this delay is of no signicance. The main objective of the simulation was to examine the technical issues: ease of implementation, and to verify and conrm that the solutions do not cause performance problems. The eectiveness of the adopted

9 Average, maximal, minimal and standard deviation time of duration of 10 simulations Table 1.

avg. max. min. std. dev. Negotiation process [ms] 3 917.6 4 225.0 3 726.0 149.4 Relevant negotiation process [ms] 933.6 1 223.0 754.0 138.6 Single iteration [ms] 50.8 133.0 30 16.2

multilateral negotiations scenario in the concerned case of a simple but realistic example of the problem of trade, has been conrmed. The correctness of the protocol also has been conrmed on the example. The implementation process of multilateral negotiations in the proposed architecture and using the selected tools was relatively simple. The use of M3-XML dialect allows to easily formulate subsequent negotiating proposals.

6

Summary

The method for reaching agreements using the multilateral negotiations in a distributed environment, using software agents, was proposed. As an interesting trade problems that require multilateral transactions, may aect the multi-

3

commodity trade, the application of M model was also suggested. The paper presents a simple, but realistic case study of a problem whose solution requires the use of the multi-commodity trading mechanism and the multilateral negotiations. Also, results of simulation of the negotiations were presented and analyzed. Of course, the case study described herein relate to a simple example, but it helped to determine the usefulness of the proposed approach: multilateral negotiations in electronic, multi-commodity trade. Nonetheless, despite the simplicity of the example, such situations may occur in the actual trading. In the real multilateral negotiation, the participants' concession strategies, as set forth in [3], should be taken into account. Thus our further work concerns the inuence of concession strategies on the properties of the protocol for the multilateral negotiation. Also we will compare our negotiation protocol with the protocols existing in the literature.

Acknowledgments The research was supported by the Polish National Budget Funds 2009-2011 for science under the grant N N516 3757 36

References 1. Adams, G., Rausser, G., Simon, L.: Modelling multilateral negotiations: An application to california water policy. Journal of Economic Behavior & Organization 30(1), 97111 (1996)

10 2. Bartolini, C., Preist, C., Jennings, N.: A software framework for automated negotiation. In: Choren, R., Garcia, A., Lucena, C., Romanovsky, A. (eds.) Software Engineering for Multi-Agent Systems III, Lecture Notes in Computer Science, vol. 3390, pp. 213235. Springer Berlin / Heidelberg (2005) 3. Endriss, U.: Monotonic concession protocols for multilateral negotiation. In: Proceedings of the fth international joint conference on Autonomous agents and multiagent systems. pp. 392399. AAMAS '06, ACM, New York, NY, USA (2006) 4. Foundation for Intelligent Physical Agents. http://pa.org/ 5. Kacprzak, P., Kaleta, M., Paªka, P., Smolira, K., Toczyªowski, E., Traczyk, T.: Modeling distributed multilateral markets using Multi-commodity Market Model. In: Information Systems Architecture and Technology: Decision Making Models, pp. 1522. OWPW, Wrocªaw (2007) 6. Kacprzak, P., Kaleta, M., Paªka, P., Smolira, K., Toczyªowski, E., Traczyk, T.: Communication model for M3  Open Multi-commodity Market Data Model. In: 2nd National Scientic Conference on Data Processing Technologies KKNTPD'2007. pp. 139150 (2007) 7. Kaleta, M., Paªka, P., Toczyªowski, E., Traczyk, T.: Electronic trading on electricity markets within a multi-agent framework. In: Computational Collective Intelligence. Semantic Web, Social Networks and Multiagent Systems, Lecture Notes in Computer Science, vol. 5796/2009, pp. 788799. Springer Berlin / Heidelberg (2009) 8. McLaughlin, B.: Java and XML data binding. O'Reilly & Associates Inc. (2002) 9. Nimis, J., Lockemann, P.: Robust Multi-Agent Systems The Transactional Conversation Approach. In: SASEMAS'04  Third International Conference on Autonomous Agents and Multi-Agent Systems. New York (2004) 10. Oberthur, S., Ott, H.: The Kyoto Protocol: international climate policy for the 21st century. Springer (1999) 11. Paªka, P., Caªka, M., Kaleta, M., Toczyªowski, E., Traczyk, T.: Design and java implementation of the multi-agent platform for multi-commodity exchange. In: 3rd National Scientic Conference on Data Processing Technologies KKNTPD'2010, pp. 184196. OWPW (2010) 12. Paªka, P., Traczyk, T., Wilk, R.: Centralized multilateral negotiations in multicommodity trade using the M3 and ebXML standards (in polish). In: Studia Informatica, vol. 32, pp. 455466. Wydawnictwo Politechniki ‘laskiej, Gliwice (2011) 13. Paªka, P., Kaleta, M., Toczyªowski, E., Traczyk, T.: Use of the FIPA standard for M3  open multi-commodity market model (in polish). In: Studia Informatica, vol. 30, pp. 127140. Wydawnictwo Politechniki ‘laskiej, Gliwice (2009) 14. Rausser, G.C., Simon, L.K.: A noncooperative model of collective decision making: A multilateral bargaining approach. Tech. rep., Department of Agricultural & Resource Economics, UC Berkeley (Aug 1992) 15. Rosenschein, J.S., Zlotkin, G.: Rules of Encounter. MIT Press (1994) 16. Simon, L., Goodhue, R.E., Rausser, G., Thoyer, S., Morardet, S., Rio, P.: Structure and Power in Multilateral Negotiations: An Application to French Water Policy. No. 37630, University of California, Davis, Giannini Foundation (2007) 17. Wooldridge, M., Bussmann, S., Klosterberg, M.: Production sequencing as negotiation. In: In Proceedings of the First International Conference on the Practical Application of Intelligent Agents and Multi-Agent Technology (PAAM-96. pp. 709 726 (1996) 18. Woolridge, M.: Introduction to multiagent systems. John Wiley & Sons (2001) 19. Zeuthen, F.: Problems of monopoly and economic warfare. G. Routledge & sons, ltd. (1930)

Suggest Documents