The Dynamics of Computing Agent Systems

Abstract

Abstract. The paper presents the Multi Agent System (MAS) designed for the large scale parallel computations. The special kind of diffusionbased scheduling enables to decompose and allocate the migrable computing agents basing only of the local information. The paper introduces the formal model of the MAS under consideration in order to depict the roles of agent behavior and the whole system dynamics. The optimal scheduling problem for MAS as well as the way of its verification was presented in terms of such model. The brief report of the test results is stressed in the section 6. MotivationThe Multi-Agent System (MAS) governed by the local diffusion scheduling is a reasonable alternative for the centrally governed distributed computing systems. Although the idea of mobile task diffusion is well known since more then ten years (see. e.g. [6]) we propose the policy that consists in on-demand task partitioning and task remaping obtained by the dynamic agent creation and migration. It was implemented and initially tested for the regularly and irregularly concurrent computations (see [2,1,5,7]). The brief report of the test results is stressed in the section 6. The paper introduces the formal model of the MAS under consideration in order to depict the roles of agent behavior and the whole system dynamics. The optimal scheduling problem for MAS as well as the way of its verification was presented in terms of such model. Formal Description of the ArchitectureThe MAS under consideration that allows the diffusion governed scheduling is a collection of: a computational environment (MAS platform) and a computing application composed of mobile agents called Smart Solid Agents (SSA). The computational environment is a triple (N, B H , perf), where: N = {P 1 , . . . , P n } , where P i is the Virtual Computation Node (VCN). Each VCN can maintain more than one agent (the number of hardware processors usage is not relevant in our assumptions).

Task Hibernation in a Formal Model of Agent-Oriented Computing Systems

“…Let us now recall (after [7]) the definition of the optimal scheduling for a computing MAS in terms of the stochastic optimal control theory. We have already the state equations (11), so we need also a cost functional and a set of admissible controls.…”

Computing MAS Dynamics Considering the Background Load

“…[2,5,6]). The agent migration and partitioning rules described in these papers are related to the phenomenon of the molecular diffusion in crystals.…”

“…The agent migration and partitioning rules described in these papers are related to the phenomenon of the molecular diffusion in crystals. The agent A i migrates along the gradient of its binding energy which depends mainly on its remaining computation time E i and the local load concentration [2].…”

“…-It delivers the information about the local load concentration L j and Q j (see (2) and (3) below), -It performs the agent migration between the neighboring VCN's, -It executes the agent partitioning and destruction, -It supports the transparent communication among agents.…”

“…[1]). A formal model of such systems was introduced in our earlier papers [2,3,4]. It enables us to formulate a new definition of the optimal scheduling problem which is well suited to the diffusiontype distributed government of computing agents.…”