![](bg3.png)
3.1 Agent Based Peer Manager
[APM]
The Agent based Peer manager is an intelligent
agent that handles negotiating and registering of
services based on directory services. Although it
poses a threat as a centralized scheme, it’s primarily
used for discovery and communications between
agents and their respective peers based on a light
weight directory services model such as LDAP
(Source: Open LDAP directory service protocol -
Source: www.openldap.org). The primary role of the
APM is to register services of agents based on
commonality and self-interests factors and also help
in discovery and formation of Coalitions in agents
based on the commonality of their services rendered.
The APM also tends to act as a broker or
middleware for all agents associated with it. An
economic brokering system could be incorporated
based on the usage of the APM which renders as a
service to self-interested agents and agent based
systems in its locality.
3.2 Electing a RSD Leader
Let there be a set of agents {A, B...E} located in
nearby locations. We assume that these agents form
a Regional service domain [RSD] based on a set of
attributes that are adhered to by the agents and the
service requestor. That is all the agents available in a
common region are grouped by a leader called the
Regional leader. So each region can have a number
of RSD’s based on the how close agents are to each
other. Once the RSD is formed each agent votes for
a leader for representing the agents in a particular
region. Here the agents say {A, B, C, D, E} will bid
on the Job J1 along with A the elected leader by
forming new coalitions. The least loaded Coalition
having an optimal turn around time will get the job
based on the negotiated payoff value. R after
negotiations decides upon selecting the best
coalition for the job and offloads the job for job
processing. The primary use of the RSD leader is to
represent a set of closely-knit agents to minimize the
communication costs between the agents and the
APM.
4 COALITION FORMATION
METHODOLOGY
This architecture allows agents to dynamically form
coalitions in order to bid for, executing tasks. Firstly
the job description will be broadcast to all agents of
the RSD. Agents exchange information with respect
to their present ability to perform the tasks. The
information is composed of a set of attributes that
explains the current state of the agents. There is a
utility function that assigns a utility to the coalition.
The utility function takes the attributes of all the
coalition members and computes a utility that
indicates which coalitions are suitable for the tasks.
Agents then try to form coalitions based on this
indicative utility. Agents decide to join a coalition if
they are offered a payoff, which is not less than a
payoff threshold computed individually.
4.1 Ranking by Indicative Utility
The information is a set of attributes or properties
that affect the way jobs are to be processed by
agents and their respective peers. Attributes in
agent-based grids can be the load of machine, turn
around time of a process, latency, QoS factor,
bandwidth requirement, distance, etc. Each agent
maintains an attribute table, which contains all other
agents attributes it collects after the exchanging of
information. Based on the table shown in Figure 4
we can assume that each job is associated with an
attributed value with respect to the properties
associated by that attribute. Based on this we can
compute the utility value for each agent by. The
figure 4 indicates that Agents A, B and C have A1
…. A3 attributes which are needed for computing
the utility value of each potential coalition based on
the attributes of the job. Each agent can then
compute all the possible coalitions it may form.
These coalitions are ones that each agent would like
to propose to other agents. The agent uses each
coalition in this list as a proposal for forming
potential coalitions. It then computes the utility
value for each coalition. The utility value indicates
how well a coalition can perform a task. The utility
can be computed by applying a LESS or a MORE
value with respect to the properties of the attribute.
For example if we were computing the Load of
individual agents then its useful to have a less loaded
system than that of the jobs attribute value. So a
coalition formation will lead to an average of loads
to compute the average value of the potential
coalition and a LESS is specified for computing the
validity of the potential coalition formation between
say agents A and B. The agent then ranks coalitions
based on their utility values in a ranking table called
RTable in descending order. In the case that the
multiple coalitions yield the same utility value, then
smaller coalitions are preferred. Based on the
ranking table RTable the best coalition formation
strategy is deployed. The price for executing the job
successfully will be equally distributed among the
coalition members as their individual payoffs.
ICEIS 2005 - SOFTWARE AGENTS AND INTERNET COMPUTING
172