What are the current R&D Directions in Mobile Agents?
- Performance
- Choosing when to move
- Choosing where to move
- Scaling to small platforms
Performance
Tests can easily demonstrate that there can be a reduction of up to 50% in the time it takes for a multi-agent system to resolve a problem using the concept of agent meeting places. In a wireless network this savings can be up to 75%.
This savings comes primarily from the overhead in TCP/IP communications and network performance.
Choosing when to move
An agent should jump when:
- the local environment has become suboptimal;
- the local environment will cease to exist.
An agent will jump to a new machine, M, if:
- M has unique resources required by the agent;
- M has the best environment for the agent’s next task(s).
In a dynamic environment, this requires "planning" or "control". Solving some sort of optimization problem. Planning, in turn, requires knowledge of "state."
- NOT agents for planning BUT planning for agents…..
Problem: Formulate the decision to jump and the decision where to jump as an optimization problem. This is generically NP-Hard because it involves TSP type problems, uncertainty about future environments, decisions about whether to clone or not, etc.
Where to go?
Depends on :
- locations of candidate services;
- network latencies;
- network bandwidths;
- machine loads.
This requires an infrastructure for supplying this type of information.
[MobileAgentPlatforms | MobileAgents | End]
(last edited November 25, 1999)
Find Page by browsing or searching
MobileAgentResearch