A Reliable Message Delivery Protocol for Mobile Agents M.Ranganathan, Marc Bednarek and Doug Montgomery Internetworking Technologies Group National Institute of Standards and Technology 100 Bureau Drive, Gaithersburg, MD 20899. mranga, bednarek, dougm @antd.nist.gov Abstract. The abstractions and protocol mechanisms that form the basis for inter-agent communications can significantly impact the overall design and effectiveness of Mobile Agent systems. We present the design and performance analysis of a reliable communication mechanism for Mobile Agent systems. Our protocols are presented in the context of a Mobile Agent system called AGNI . We have developed AGNI communication mechanisms that offer reliable peer-to-peer communications, and that are integrated with our agent location tracking infrastructure to enable efficient, failure-resistant networking among highly mobile systems. We have analyzed the design parameters of our protocols using an in-situ simulation approach with validation through measurement of our prototype implementation in real distributed systems. Our system assumptions are simple and general enough to make our results applicable to other Agent systems that may adopt our protocols and/or design principles. 1 Introduction Mobile Agents are a convenient and powerful paradigm for structuring distributed systems. Using the Agent paradigm, work can be assigned to sequential, event-driven tasks that cooperate with each other to solve a dis- tributed problem. In such systems, Agents roam the network accessing distributed information and resources while solving pieces of the problem. During the course of these computations Mobile Agents need to communi- cate among themselves to exchange state and status information, control and direct future behavior, and report results. The abstractions and protocol mechanisms that form the basis for inter-agent communications can significantly impact the overall design and effectiveness of Mobile Agent systems. Numerous approaches to inter-Agent communications are possible including RPC and mailboxes [3]. In this work we present a simple, ordered, reliable, one-way message protocol on top of which other abstractions can easily be built. Reliable, ordered one-way communication mechanisms greatly simplify the construction of most distributed applications. In traditional distributed applications, TCP [15] provides such services. Through decades of expe- rience and re-engineering, TCP has evolved into a protocol that is highly effective at providing reliable end-to- end data delivery over the conditions found in today’s Internet (e.g., link failures, variable latencies, congestion loss). In this paper, we examine how to build a TCP-like reliable communication mechanism for Mobile Agent sys- tems. The first question to be addressed is ”Why we don’t use the existing TCP?” We argue that Mobile Agent systems impose new communication requirements and problems that are not adequately addressed by con- ventional TCP, nor its potential minor variants. In particular, we are concerned with building failure resistant, rapidly re-configurable distributed systems. We view these system properties as a primary motivation for dy- namic Agent creation and mobility mechanisms, and as posing significant requirements on inter-Agent commu- nications mechanisms. As such, we require that Agent systems be able to (1) detect and recover from failures in the end-to-end transport mechanism and (2) accommodate efficient communication among mobile end-points. Neither of these capabilities can be provided using standard TCP. In the remainder of this paper we present the design and performance analysis of a reliable communication mechanism for Mobile Agent systems. Our protocol is presented in the context of a Mobile Agent system called AGNI, whose general design and capabilities have been described earlier [13]. Our communication mechanism AGNI stands for Agents at NIST and is also Sanskrit for fire. This work was supported in part by DARPA under the Autonomous Negotiation Teams (ANTS) program (AO # 99-H412/00). The work described in this paper is a research project and not an official US. Government endorsement of any product or protocol.