Cooperative Agent Learning Model in Multi-cluster Grid
Qingkui Chen, Songlin Zhuang, He Jia XiaoDong Ding · InTech eBooks · 2011
With the rapid development of the information techniques and Internet, and their popular application, the research results based on CSCW became the key techniques to build the enterprise information infrastructure (Raybourn & Newman, 2002).The computer based on cooperative work environments is playing the more and more important role in the business behavior of the enterprise today.Especially, the CSCW contains a lot of computation-intensive tasks and they need to be processed in some high performance computes.On the other hand, Intranet is more and more extended and a great number of cheap personal computers are distributed everywhere, but the utilization rate of their resources is very low.The researches of papers (Markatos & Dramitions, 1996) (Acharya&Setia, 1998) point out that many resources are idle in most network environments at a certain time.Even it is the busiest time of a day, still one third of their workstations aren't used completely.So, the paper (Chen & Na, 2006) proposed a framework which is how to collect and use the idle computing resources of CSCW environment that composed of multi-clusters connected by Intranet.It uses the idle computing resources in CSCW environment to construct a Visual Computational System (VCE).VCE can support two kinds of migration computations: (1) the Serial Task based on Migration (STM); (2) the Task based on Data Parallel Computation (TDPC).For adapting these heterogeneous and dynamic environments, we use the Grid (Foster & Kesselman, 1999) techniques and multi-agent (Osawa, 1993) (Wooldridge, 2002) techniques, and collect the idle resources of CSCW environment to construct multi-cluster grid (MCG).Because the migration of computing task and the dynamic changes of the idle resources in MCG, the intelligence of computing agents for raising the utilization rate of idle resource becomes very important.So, the effective learning model of agents is the key technique in multi-cluster grid.There are a lot of researches about agent learning (Kaelbling et al., 1996) ( Sutton, 1988)( Watkins & Dayan,1992) ( Rummery&Niranjan, 1994)( Horiuchi&Katai, 1999) nowadays.It includes two aspects: passive learning and active learning.The main theory of active learning is the reinforcement learning.At the same time, the agent organization learning (Zambonelli et al., 2000) (Zambonelli et al., 2001) model become another focus.The problem of distributed and cooperative multiagent learning is studied through complex fuzzy theory in the paper (Berenji&Vengerov, www.intechopen.com