ABSTRACT
Basically each peer can be classified into two types of peer such as a group peer (GP) and a client peer (CP). First, GP has many responsibilities to manage a group and peers and to communicate with other GPs as well. And it has a lot of functionalities as following: control message processing, resource discovery (e.g. Grid services, data, and computing element), and store the metadata of peers with index. In addition, as aspect of Grid middleware, it should take a role of the quality of services negotiator, job executor with monitoring, and aggregation of result sets. On the other hands, the latter can be considered as a resource consumer and a resource provider at the same time in the proposed system. It is obvious that GP will experience a problem in terms of system over-load, namely failure of single point. As the alternative way to over come this, we propose the redundant GPs scheme for dependability and performance reasons. This computing model can be modeled by the k-level of a redundancy. Simply we consider that increasing k value makes system more reliable than system with low level of redundancy. However, one more thing to consider next is that intuitively, unlimited duplication is improper due to infinite cost. Thus, redundancy optimization policy regarding performance issue (e.g. cost-wise strategy) should be discussed in this paper.
