Agent computing is subject to several inefficiencies, some of which are due to the complexities of the environments in which mobile agents are deployed. Our research is exploring runtime adaptation and agent specialization to improve the performance of agent-based programs, aiming at enabling programmers to employ these techniques and runtime adaptation in general, to improve program performance without sacrificing the fundamental advantages promised by mobile agent programming. We explore the effects of using two specialization approaches, morphing and fusion, on a single mobile agent and on several cooperating agents. Our experimental results with two sample applications, ISDA and PSSPS, show that such specialization approaches result in considerable performance improvement.
We have built a preliminary infrastructure for online morphing, which offers mechanism for inter-language remote invocation using the model of invocation adaptors developed in our research. Infrastructures and mechanisms are applied to the ISDA and PSSPS distributed high performance applications. Also used with these applications is a realization of mobile event channels that allow reliable event delivery during end-point migration.
Our future work concerns systematic support for specialization approaches like morphing, fusion and others such as slicing. This support will comprise event mechanisms and quality of service infrastructure, both of which are important to a general agent adaptation system. We will also work on compilers for agent fusion and the adaptation of agent invocations.
Acknowledgments. Prof. Raja Das is participating in the Objent system design, with focus on the application of compilation methods to improve agent or object performance, including the use of fusion techniques. Prof. Mustaque Ahamad has been investigating the overheads of agent communications (e.g., Java RMI) and methods for creating quality-controlled Java-based objects. Greg Eisenhauer and Beth Plale are responsible for the ISDA application, Davis King provided the code for Isosurface calculation, and Fabian Bustamante is currently creating additional elements part of the ISDA application. Rajkumar Krishnamurthy provided the SAR benchmark code used in our work.