Designing an information system for Grids: Comparing hierarchical, decentralized P2P and super-peer models

As deployed Grids increase from 10s to 1000s of nodes, the construction of an efficient and scalable information system is a key issue, as it is vital for providing querying and discovery services. Today most Grids adopt a centralized or hierarchical model for their information system, but this model is characterized by poor scalability, resiliency and load-balancing features. Nowadays the research and development community is heading towards the use of scalable information systems based on distributed models such as the decentralized peer-to-peer (P2P) model and the super-peer model. If the former is adopted, each node can act both as a client and a server as it can generate discovery requests and also respond to requests issued by other peers. Requests and responses are forwarded with a hop-by-hop mechanism by ad hoc Grid Services hosted by Grid nodes. The super-peer model is a recently proposed approach that combines features of centralized and P2P models. A super-peer acts as a server for a single Grid organization, and publishes metadata describing the resources provided by the nodes of that organization. At the same time, super-peers connect to each other to form a P2P network at a higher level. This paper analyzes information systems based on three alternative models: the hierarchical, the decentralized P2P, and the super-peer model. A performance evaluation of such models is reported, and afterwards a performance comparison is discussed in order to analyze the pros and cons of each solution.