A multi-agent based autonomic management architecture for large-scale server clusters

As the size of the server cluster increases, management of clusters has been a daunting task. Traditional management architecture of cluster system fails in providing efficient management as the scale of a cluster increases. In this paper, a novel management architecture of a large-scale cluster system, called cooperation-oriented hierarchical autonomic architecture (CHAA), has been proposed to achieve an efficient management. We employed multi-agents to construct CHAA. Considering that the intrusion caused by bidding increases dramatically when the number of agents is large, we proposed hierarchical Contract Net Protocol (HCNP) to relieve the overhead. Generalized partial global planning (GPGP) has been introduced into CHAA to enable agents recognize their potential interrelationship and it has been combined with HCNP as the cooperation method of agents in CHAA. We made performance measurement on the traditional management architecture and CHAA, and the results showed that CHAA is a light weight management architecture which promises good scalability.