|
Project: Resource Management
|
Introduction
The aim of this research is to contribute to the development of AgentScape: middleware for
the development of large-scale multi-agent systems. The current focus is on determining requirements
for, and characteristics of, a scalable resource management system for large-scale multi-agent
systems. This includes the application of resource management in existing agent platforms and
other related areas of research, e.g. distributed computing. A prototype of a method for resource
management is to be developed.
Large-scale distributed systems are often heterogeneous systems : heterogeneous with respect
to the host architecture (Sun SPARC, Intel x86/i64), the supported operating system (Solaris,
Linux, Windows NT), and the communication infrastructure (bandwidth and latency). A mobile
agent is a migrating process that can autonomously decide to move from one machine to another.
An important characteristic of agents, which distinguishes them from traditional processes, is
autonomy; agents are in control of their own behaviour. From a management point-of-view, this
has implications for the management of these mobile autonomous processes.A multi-agent system
management architecture should have the ability to regulate agents without interfering with
their autonomy.
A multi-agent system needs to manage agents, objects with which agents interact, as well as provide access to external services. The
OSI management model defines five functional management areas: Fault-tolerance, Configuration,
Accounting, Performance and Security. These areas can also be applied to the domain of multi-agent
system management:
- Performance management: managing the resource usage of agents on hosts within a multi-agent system.
- Security management: ensuring security of both agents and hosts in a multi-agent system.
- Fault-tolerance management: ensuring the availability of agents and objects within a multi-agent
system, as well as the multi-agent system as a whole.
- Account management: maintaining administrative information about the agents within a system.
- Configuration management: managing the configuration of agents and objects that are under management control.
A management system in an agent operating system needs to create, delete, migrate, etc.
agents. An agent life cycle model forms the basis for the definition of these operations on
agents by defining states of the agent and transitions between these states. The central state
in the life cycle model is the suspended state, in which the agent is inactive and can be
manipulated. This is in contrast to other life cycle models for agents, in which the active
state of an agent is the central state.
Current developments
Master student Michiel Meijers has recently finished his thesis entitled: "Agent-Based Document Assembly
Framework: Examining Agent System Resource Management in a Real World Scenario". The project focused on
creating a flexible agent-based alternative to current server-based document assembly systems. As the project is
part of the Resource Management research within the AgentScape project, an important goal was to enable proper
management of access to assembly related resources by agents.
This research on management of mobile agents in a large-scale distributed system is part of
our research on a large-scale scalable distributed agent operating system. See our main
Research Projects page for more information on other IIDS projects.
This research project is financially supported by Stichting NLnet.
