Software Agents in ADAJ: Load Balancing in a Distributed Environment

Abstract

Adaptive Distributed Applications in Java (ADAJ ) is a platform developed for execution of distributed applications in Java. The objectives of this platform is to facilitate application design and to efficiently use the power of distributed computing. The ADAJ offers both a programming and an execution environment. In the latter it implements object observation and load balancing mechanisms. The observation mechanism allows estimating of the JVM load for each node running the ADAJ client. The load balancing mechanism dynamically adapts the workload across the system according to this information. Here we discuss how the original design based on JavaParty is going to be superseded by utilization of software agents.

Combining a JADE-Agent-Based Grid Infrastructure with the Globus Middleware Initial Solution

“…Separately, in [19] we have discussed trust-management-related issues. Finally, in [14] we have contemplated how the proposed approach can be intermixed with work done within the ADAJ project (see, also [23]). …”

Resource Management in Grids: Overview and a discussion of a possible approach for an Agent-Based Middleware

“…etc. ), (c) upon receiving such list, the LAgent eliminates these that are not trustworthy [38], (d) next it utilizes the FIPA Contract Net Protocol [13] to negotiate which team to join (note that, in the context of this chapter, we omit details of contract negotiations; e.g. how the LMaster establishes the optimal price; we simply assume that one of bargaining mechanisms described in Section 3.1 is used), (e) negotiations can result in a success (joining a team), or in a failure (no acceptable team was found).…”

“…The second aspect that needs to be considered is: what happens when User would like a job to be executed in our system. As described in [34] the process consists of the following steps: (a) User specifies to its LAgent acceptable conditions of job execution, (b) the LAgent contacts the CIC agent to obtain the list of teams that have the required resources, (c) this list is then adjusted on the basis of trust considerations ( [38]), (d) as in the previous case, FIPA Contract Net protocol is utilized as a negotiation mechanism to find the best team to execute the task, (e) process can result in a success (finding a team to do the job), or in a failure (no acceptable team is found). Here, of particular interest is available information concerning the job.…”

“…As a matter of fact, they should support each other, as the success of the team means also their (financial) success. Obviously, this assumption may need to be changed (introducing agent spoilers and/or competitive/selfish behaviors of Workers), but this would require reexamination of all trust-focused considerations (see [38]) and thus, for the time being, will not be pursued further. In the negotiation-based approach, when the LMaster wins a contract, it could utilize the FIPA Contract Net to inform Workers and to find out which of them could execute it.…”

Ontologies, Agents and the Grid: An Overview

“…It is assumed that all information in the system will be ontologically demarcated and semantically processed. Summary of the results obtained thus far can be found in [18,23,31,17], while the current version of the code is available at [1].…”