- Xuan Thang Nguyen, PhD student
-- Modeling and Solving QoS
Composition Problem Using
Fuzzy DisCS --
Multi-agent systems are a promising paradigm in distributed problem
solving and optimisation, in particular distributed
constraint satisfaction problems (DisCSP). We discuss different
agent-based DisCSP algorithms, including ADOPT (Asynchronous
Distributed OPTimization)
and address their shortcomings in a real environment where delay of
messages is inevitable. We present an application of agent-based DisCSP
algorithms in
the Web service domain. In particular, we investigate the problem of
QoS guarantees for multi-Web service compositions and show that this
problem can
be modeled as an instance of DisCSP and solved by cooperating agents.
We propose an agent-based Fuzzy DisCSP algorithm that solves the QoS
guarantee
problem while maximizes the minimum preference of all providers over
the final solution and maintains their different levels of importance
to the
compositions. We also present our experimental data to show the
improvement of our algorithm over other existing algorithms.
- Jian Feng Zhang, PhD
student -- Agent-based Distributed Planning--
In distributed agen-based planning (e.g. service compositions), a
single agent may not have access to all services which can make up a
composition to
fulfill a given goal. Cooperation among multiple agents, each of whom
has a limited set of services, can increase the chance of generating a
valid/effective composition. However, the agents may be reluctant to
reveal their services to other agents during cooperation due to the
concerns to
privacy, security, and difficulty of heavy communication overhead. We
developed an algorithm that allows agents to cooperate and form a
composition in such a situation. This algorithm can be viewed as a
distributed variation of the reputed planning paradigm, i.e. Graph
planning
+ CSP. We adapt classical planning graph to distributed planning graph,
represent the graph in distributed CSP (DisCSP) instead of CSP, and
solve
the DisCSP using existing DisCSP search algorithm. The agents are of
equal authority and there is no need of centralized coordinator all
through the
process.