Building research grade multi-agent systems requires a large amount of software infrastructure. Many systems require planning, scheduling, coordination, communication, transport, simulation, and module integration technologies and often the research interest pertains to only a portion of the technology or to aggregate system performance. To advance scientific progress, we, as a community, need to share this infrastructure wherever possible -- reuse in this context may enable more researchers to build functioning experimental systems and to test their theoretical ideas in actual software environments.

When these research ideas are then translated to commercial systems, scalability issues become significant. Commercial success for MAS will require scalable solutions -- in infrastructure and software design approaches, to enable re-use and effective deployment of both mobile and intelligent agents.

This workshop consists of two separate topics that are joined under the common umbrella of addressing questions that will make large scale MAS a reality. The first topic, denoted "Topic A" below, focuses on available infrastructure and requirements for constructing research-grade MAS. This topic will be the concentration of the morning session of the workshop. The second topic, denoted "Topic B" below, will aim to consider support in infrastructure and software design methods for MAS that can directly support coordination and management of large multi-agent communities. Topic B will be the concentration of the afternoon session of the workshop. A panel discussion in the afternoon, consisting of representatives from each topic area, will explore issues particularly relevant to both areas of interest.

All participants to the workshop must submit either a research statement or a full length paper to the appropriate topic organizer below. Research statements may be no more than two pages. Full length papers will be considered for presentation. Full length papers may not exceed 6 pages and should be formatted according to the Agents guidelines for the primary conference publications. All submissions are due by April 15, 2000. Acceptances/notifications will be sent via email.

Important Dates

• Papers due: April 15
• Notification to authors: May 1
• Final version of papers due: May 10

The purpose of this topic is twofold: 1) to gain an understanding of existing infrastructure for MAS, and 2) to understand how the communities' different research foci translate into requirements for the infrastructure. Accordingly, we invite papers belonging to two classes:

A) Implemented infrastructure - systems, control components, frameworks, etc., that are implemented and available (or can be made so) to outside researchers. Papers of this class may be somewhat tutorial or descriptive in nature, possibly identifying the features or components of the infrastructure and identifying the research questions the infrastructure is designed to help address/explore.

B) Research requirements or features that researchers believe are critical to their MAS research interests. Papers of this class should help researchers with infrastructure understand the relationship between their infrastructure and the needs of other researchers in the community.

The goal is to facilitate a dialogue between different infrastructure builders and between infrastructure builders and other interested members of the research community. In both paper classes, we strongly encourage pragmatic and practical submissions focused on the issue of software infrastructure for research grade MAS.

In deploying multi-agent systems for practical applications, it is important to consider infrastructure that can cater for agent and platform heterogeneity, and subsequently to large number of agents, with such agents working collectively. The infrastructure can be application dependent, or it may be possible to identify general issues that have a wider use. Emerging applications in scientific computing, e-commerce and mobile/embedded systems, where the agent paradigm is of particular significance, infrastructure scalability becomes an important consideration. Scalability can be associated with the underlying communications mechanisms, naming scheme (service), negotiation and service discovery protocols, caching strategies, and convergence time in learning agents. Scalability and performance management approaches range from analytic approaches (such as process algebras, queueing models, Petri nets) to performance engineering approaches which can be associated with software engineering methods. Also associated with the notion of scalability are aspects of robustness and persistence for checkpointing and managing state within an agent community. For instance, how can agents be moved across server boundaries without starting/stopping them? Can we dynamically identify groups (coallitions or ``congregations'') of agents in multi agent communities that could support scalability?

This part of the workshop will aim to address these issues, and long papers and short statements/research goals are invited. Your paper (long/short) should clearly identify relevance to the proposed topic area of scalability and infrastructure support for performance management.

• K. Suzanne Barber,
  University of Texas at Austin

• Tim Finin,
  University of Maryland Baltimore County

• Katia Sycara,
  Carnegie Mellon

• David Wolpert,
  NASA Ames

• Keith Decker,
  University of Delaware

If you would like to participate in this workshop, as a presenter (author) or a delegate, please send us an email.