Agents and P2P Combined For E-commerce

Electronic commerce technology is more and more present and people are getting connected together. Companies and users want to automate their task and discover new business opportunities. Many standards and initiatives have been already proposed. Most of them focus on business-tobusiness protocol definition. Rosettanet, cxml, xcbl and bizTalk are only a few examples. Users and companies then have a set of protocols they need to understand and master for implementing solutions. With the increasing number of protocols, companies need a solution permitting them to easily integrate different protocols. Unfortunately, the commercial solutions provided today do not provide sufficient support for integration. Software agents have already been proposed as a key technology for connecting people and adding new features in electronic trading. However, most of the agent-based solutions already proposed have not really changed the user’s experience, neither have they brought a major shift from traditional activities in classic markets. This is mainly due to their inflexibility. In fact, these solutions predefine rules and policies in ways that users feel uncomfortable and are limited in their actions. Agent-based systems seem to appear as a key technology that gives users more autonomy and that is yet flexible enough to integrate new business scenarios. In this paper we propose an agent-based solution that could be used to wrap in existing systems and applications. We show how it could be easy to support new protocols without being forced to rewrite solutions from scratch. The work proposed combines both, advantages from software agent technology, and from P2P networking technology. The solution has four layers where each one encapsulates a specific set of functionalities: communication, business, cooperation and coordination. Users can adapt or modify the internal structure of one layer without necessarily affecting the others. INTRODUCTION The continuous development of Internet technology and the increasing number of people getting connected to it, explains the many efforts of attempts to create virtual markets. In such a market, agents can negotiate, buy and sell goods. Those agents could either be controlled by humans or be autonomous. Many projects have proposed intelligent agent-based systems supporting many aspects of ecommerce. For example, many agent-based online shopping services have been created [1,2,3,4,5]. Despite their added value for the user shopping experience, they lack facilities for automated negotiation and agent cooperation. Other more sophisticated applications have been proposed. Kasbah [6] is a Web-based system where users can create autonomous agents to buy and sell goods. Doing this, users do not need to stay connected or watch their agents. However, Kasbah can be qualified as a closed system. It is a proprietary system, which uses a very limited message protocol. Thus, it is impossible to integrate new heterogeneous agents. MAGMA is an agent-based electronic commerce architecture, which focuses on the components of traditional markets such as communication, transfer of goods, money handling and transaction mechanisms [7]. The MAGMA architecture models traditional market activity through a platform-independent API. It uses a central communication relay server, which could be the failure point of any industrial adaptation. The traditional market and the centralized systems already proposed could be classified as classic solutions. These classic solution share the following list of limitations [8]: • Inflexibility: every e-commerce solution has it own trading environment. Unfortunately, it’s often impossible to combine concepts coming from more than one environment in the same transaction. • Only price counts: most current systems, offer auctions or negotiation to select either buyer or seller. However, they use only the price as the key point as a selection criterion. Unfortunately, this is not the main criterion for striking a deal [9]. • Centralization: trader may want to have more control. Currently, the central coordinator has the last word. He decides what to do, how to match and when. • Static predefined behavior: the proposed systems suppose that the environment is stable and predictable. Thus, they predefine the agent’s behavior, its relationship and its trading partners. • Weak automation: many operations are still not supported. Overcoming these limitations is necessary to implement a realistic solution dedicated to an open environments. Open environments are characterized by having components that are (1) autonomous (acting independently), (2) heterogeneous (designed independently), (3) of dynamic membership (joining, changing, and leaving arbitrarily) and (4) of large scale (numerous). These properties are compatible with both the agent and the peer-to-peer paradigm. In fact, peer-to-peer (P2P) networks, consisting of nodes that are peers of each other, provide a suitable paradigm for implementing dynamic trading. Nevertheless, P2P tools are not able to exchange complex data and to deal with heterogeneity, coordination and data management problems. Agents are persistent computations that can perceive, reason, act and communicate [10]. Agent technology is of interest here because it adds more intelligence over the P2P technological layer. The combination of agent and P2P technology brings more autonomy and flexibility to the proposed e-commerce system. This paper proposes an agent-based integration system. The purpose of this agent system is to enable existing applications and infrastructures to conduct intelligent interaction with users or other business systems. The idea is to have a flexible integration framework, which facilitates dynamic partner selection and supports agents’ heterogeneity. The following sections explain the P2P paradigm and present a few P2P-based applications. Then a brief overview of related work is given. After that, we present our agent architecture for open e-commerce integration. The last section discuses the future work and direction of our research. THE PEER-TO-PEER PARADIGM P2P applications have started to emerge, creating more and more hype, and taking the attention of the media. However, the P2P concept is not a new one. We can say that when two computers were first connected they formed the first P2P network. Mail servers, network news servers (NNTP), and domain name servers (DNS) operate in P2P networks. Today, many factors make P2P practical for a large number of applications. These factors include the explosion of connected devices, the rapid increase of affordable bandwidth, increase of computing power, larger storage capacities, and the proliferation of information at the edges of the network. P2P is a style of computing that makes the network interactions more symmetrical. Even though there may be centralized services (like the Napster catalog) the end user peer is a significant focus of the application. The distribution of the system reduces the dependency to network and communication failures. P2P computing takes advantage of existing desktop computing power and network connectivity in allowing economical clients to leverage their collective power to benefit the entire community. P2P networks replace the traditional centralized structure of client-server interactions with peer interactions, where each machine acts as both a client and a server in the network. Unlike the client-server architecture where a client generates a request and a sever responds to the request, peers understand both queries and responses. The peers have identical capabilities and responsibilities, and all communicate symmetrically. Napster is the good example of a monolithic centralization that causes all the P2P functionality to fail if the main server fails or is disconnected. Gnutella is the opposite because no single peer, in case of failure, will significantly affect the quality of the network. Peer-to-peer computing enables applications that are collaborative and communication-focused: it leverages available computing performance, storage, and bandwidth found on systems connected to each other in a world-wide network. Today’s best known peer-to-peer applications are Napster [11], Kazaa [12], Gnutella [13], Freenet [14] or SETI@Home [15], but various research projects have been initiated in the past few years, such as Pastry [16] and Chord [17]. Although the different P2P applications share the same notion of peer-to-peer networking, the intended usage and approach varies from application to application. Napster, Kazaa and Gnutella are primarily file-sharing applications: exchanging of files between peers. Nester’s approach to information search is traditionally client-server, while Kazaa and Gnutella adhere more to the peer-to-peer philosophy and forward information search requests to neighboring peers in the network by flooding. However, they both recently introduced super-nodes for more scalable information retrieval. Freenet is more like a distributed information storage system. It pools unused disk space across potentially hundreds of thousands of desktop computers to create a collaborative virtual file system. SETI@Home is the world’s largest distributed computing system with a capacity of an estimated 26 Teraflops/sec. If we consider that the main server is the equivalent of a very large peer, SETI could be considered a P2P system. The client peers contact the main server peer to report results and to request new data for processing. They don’t need to stay connected as the average processing time could be as much as 14 hours. Pastry is a scalable, distributed object location and routing infrastructure for wide-area peer-to-peer applications. It can be used to support a variety of peer-topeer applications, including global data storage, data sharing, group communication and naming. Chord, on the other hand, focuses on a scalable peer-to-peer lookup service to efficiently locate the node that stores a particular data item. Chord provides support for just one operation: given a key, it maps the key onto a node. The JXTA [18] project works on core network computing technology to provide a set of flexible standards that can support peer-to-peer computing. It is not an application and does not define the application itself. The focus is on creating basic mechanisms and leaving policy choices to application developers.