Self-interested and Negotiation Agents

Self-interested negotiation agents seem to be very suitable for some network manage-ment activities, especially for service managemanage-ment. Most of the papers dealing with service management refer to this kind of agents. Magedanz [Mag95] shows that these agents are adequate for the provision of high-level services in an open electronic market of telecommunications services. A good starting point for covering this trend is the FIPA Network Management and Provisioning application draft [FIP97], which defines three kinds of agents, namely the Personal Agent (PA), the Service Provider Agent (SPA), and the Network Provider Agent (NPA). The latter is responsible for the provision of the net-work resources and elements which are necessary for the service implementation. Based on these network resources, the SPA is responsible for the provision of the services with the expected quality. The PA is a kind of Personal Digital Assistant [Mae94] that assists the user in defining his requirements for the application needs with regard to the user’s preferences. It then has to negotiate these requirements with different SPAs in order to find out the best provider with the best service in terms of maximum quality and min-imum cost. When a connection is about to be established, the PA has to commit the local resources of the user and to configure his equipment according to the established connection.

The SPA has to catch the user requirements, and to identify the necessary services and to map the user requirements into these services parameters. It then negotiates with the NPAs to select the best network provision, again in terms of maximum quality with minimum cost. Feedback with the PA may occur in order to conclude with the best ar-rangement both with the network provider and the user.

Finally, the NPA gets the SPA’s specifications and translates them into network

re-quirements in terms of bandwidth, jitter, etc. It may also have to negotiate with other NPAs, mostly in the case that multiple network domains are involved.

Most of the work based on agent negotiation makes use of such scenario (or simi-lar ones) in which the interacting parties include Customer Agents and Service Agents.

According to [HB99], Busuioc’s work was among the earliest to use agents in telecom-munications problems. [Bus96] describes a flat agent architecture for the management of services in a mixed mobile-fixed network. The Contract Net Protocol is used as a negotiation-based trading mechanism to enable agent cooperation. Agents involved in this architecture are Customer Agents (PA in the FIPA terminology), which ask for the es-tablishment of services from Service Agents (i.e. SPA in the FIPA terminology). Service Agents may themselves negotiate among each other according to their attributed roles and domains.

Calisti et al. [CWF99] proposes a software architecture that integrates FIPA scenario with existing TINA and TMN frameworks. In [CF99b], FIPA compliant agents are also suggested to automate the negotiation between multiple NPAs in order to provide a dynamic solution for the inter-domain demand allocation (IDA) problem. IDA con-sists of the establishement of cross-domain connectivity with specified QoS. The net-work provider internet-working (NPI) paradigm [CF99a, CFF99] relies on a distributed set of agents that model different network domains. These agents apply distributed constrain satisfaction in order to provide an efficient solution to the IDA probem.

Ruzzo and Utting [RU95] introduce User Agents to define customized services. New services can be customized by supplying the agent with the user policy. The UA is then able to make the best choice between what can be achieved as suggested by the fall-back of the called agent, and what the service provider offers. The caller and the called agents may also negotiate in order to reach an agreement that satisfies both users policies, i.e.

preferences and constraints. To support user mobility, end-point agents are also intro-duced. Mobile-user agents have to negotiate with end-user agents to get the permission to access the corresponding end-points and establish communications.

Another similar work by Griffeth and Velthuijsen [GV93] uses negotiating agents to handle feature interaction problems in value-added telecommunication services [GV94].

Other scenarios suggest implementing new services in agents. [HJ97] suggests that while services are ucustomized by PAs, new services can be created within ser-vice agents in a serser-vice-level agent environment. Serser-vice agents encapsulate network specifics and each one is dedicated to one service that can be constructed over other ser-vice agents. Similarly, [MRK96] presents a scenario where MAs are sent into an agent environment on the services host, where they can offer new services by using the legacy

services and/or by exploiting other mobile agents’ facilities. In both scenarios, the agents that have to provide new services have to negotiate with the already existing agents in or-der to come out with an agreement to implement these new service.

Negotiation is also adopted in HYBRID (Section 3.4.3) for managing ATM networks.

Agents are organized within a hierarchy of authorities. Each authority is responsible for some delegated resources and exports “performance indices” [Som96b]. When an authority (or precisely, a service agent inside the authority) receives a user request with a certain Service Level Agreement, it commits the necessary directly managed re-sources and then decides which of the sub-authorities is best eligible to route the request through. This decision is based on atendering processbased on the performance indices declared by sub-authorities. Moreover, anegotiation processis initiated between the ser-vice agent and thecustomer agenteach time the SLA parameters for a given service need to be refined, e.g. when a network failure occurs. The tendering process is supported by a set of four performatives: PROPOSE,COUNTER-PROPOSE,ACCEPT, andREJECT. The ne-gotiation process also uses four performative: PROBLEM,COUNTER-SOLUTION,ACCEPT, and REJECT. Both processes are based on finite-state machine modeled dialogues layered on top of the extended KQML.

The approach adopted in [HD98] uses negotiation for bandwidth allocation of Vir-tual Path Connections in ATM networks. The idea is to have agents associated to physical links that distribute bandwidth to agents associated to VPCs using those links. (This is similar to the model proposed by Hayzeldan et al. [HBL99, Hay99, HB98a].) Agent ne-gotiation is based on a communication language similar to FIPA ACL. In the case that a VPC is congested, links agents (i.e. agent associated to physical links) have to agree on the amount of bandwidth to be allocated. This amount cannot exceed the spare band-width on the most used link. This negotiation is based onINFORMmessages followed by a REQUESTmessage from the link agent with less spare bandwidth. In the case that there is a link with no spare bandwidth to be reallocated for the congested VPC, one of the VPCs on this link has to be rerouted through other links. Again, negotiation takes place between the VPC agents (i.e. agent associated to VPCs) in order to determine the VPC with the most allocated bandwidth. In this case bidding is started between the VPC agents using thePROPOSEperformative, i.e. each agent broadcasts aPROPOSEmessage containing its bid. The agent responsible for the VPC with the highest amount of allocated bandwidth will produce the highest bid and will therefore commit to rerouting.

It can be noticed that negotiation within a group of agents is costly especially when multiple stage negotiation is allowed. This is due to the negotiation messages, which have to be broadcasted to all the agents participating in the negotiation. Such problem should be carefully handled as it might lead to scalability and synchronization issues.