In order to make these problem solving agent applications possible we need to understand how to deploy agents on a large scale. There are many aspects of this problem, including mechanisms to support large numbers of agents, and how a large number of agents function. Agent deployment must be sufficiently robust and reliable, so that scientists and commercial organizations will entrust agents with mission critical applications.

This workshop will be organized around two themes:

• The application of agents to problem solving applications, such as scientific computing research or large scale business problems
• Examination of the practical issues of deploying such problem solving systems.

The workshop includes, but is not restricted to, the following general areas:

Practical Deployment

• Performance analysis/modeling of multi-agent systems
• Performance enhancement methodologies for mobile and multi-agent systems
• Agent communication over MPI/PVM
• Agents in problem solving environments
• Recommender agents for scientific problem solving
• Computational steering using agents
• Software updates and component distribution via mobile agents
• Agent based load balancing
• Agents based network characterisation for high performance computing
• Agent based resource discovery
• Agent based real time multimedia and embedded systems

Communication

• How do agent communication mechanisms scale to support many agents?
• How does the agent's naming scheme scale to support a huge number of systems, services, and agents? How does the name-resolution mechanism scale?
• What is the impact of communication processing on network load when many distant agents are communicating?
• How reliable does communication have to be? What support is needed for network disconnection, due to Internet failures or due to purposeful disconnection of the host machine?
• How do the negotiation protocols that lie at the heart of ACL (Agent Communications Languages) designs scale? What role can strategies such as caching and adaptive searching play in cutting down the time needed to perform common actions, such as converge on a common ontology.
• An agent can perform actions in other locations in several ways. It can migrate there, and perform the action. It can communicate with an agent located in the appropriate locale, and ask that the other agent to perform the action. The agent can remotely invoke code in the other locale. Each of these has different costs associated with it, such as speed, or network traffic. There are other variables, such as agent size, required reliability and state, and access permissions. What are the kinds of research we need to know which model is best for various different usage scenarios?

• How can we move or update agents or agent servers without starting and stopping them?
• Given the possibility of thousands of agents running on a single server, what are the mechanisms we need to support agent persistence, such as reactivate-on-event and checkpointing? For large (possibly intelligent) agents, how can we efficiently handle the persistence of long-term data?

Implementations in particular scientific computing domains will also be considered, and submission of work-in-progress, or work completed, is encouraged. Possible application domains include:

• Molecular dynamics
• BioInformatics and protein sequence processing
• Fluid dynamics
• Collaborative virtual reality for science and engineering
• Climate modeling
• Computational cosmology
• Geological modeling

• Financial forecasting over large data sets
• Agent based scalable data warehousing
• Electronic commerce
• Enterprise wide intranet applications

Electronic submission is preferred. Please email a PostScript or PDF copy of your submission to Omer Rana (omer@cs.cf.ac.uk) before February 15, 1999. You may also send paper copies to Omer Rana, Department of Computer Science, Cardiff University, PO Box916, Cardiff CF2 3XF, UK or Kate Stout (Kate.Stout@sun.com), Sun Microsystems, 2 Elizabeth Drive, Chelmsford, MA 02124, USA.

Papers will be posted on the workshop website prior to the workshop, to allow attendees to read materials before the workshop.

Submission Deadline: February 15, 1999

Towards the end of the workshop, the general issues generated from the workshop will be examined.