Design of a Peer-to-Peer Information System Using the GT4 Index Service

Applications running on the grid need efficient access to information about resources and services. This information can significantly impact scheduling, replica selection, and planning for workflow execution. Grid information services need to be able to aggregate information from multiple types of resources and support both explicit and attribute-based queries. As grids grow larger, organizing these services into efficient hierarchies, while taking care to avoid cycles and ensure that nodes do not become hot spots, is challenging. We present a peer-to-peer organization of a distributed grid information service, the GT4 Index Service. In our design, modified GT4 indexes called P2P indexes organize themselves into an unstructured P2P system. Queries from a user to any index may be forwarded to other indexes, and query results are routed back to the user. We use the P2P network only for self-organization and query forwarding among P2P indexes, which are optimized to store resource information and process associated queries. Most previous work in unstructured P2P networks has been on file sharing applications or data storage systems. Here, we explore the issues of organizing information services into an unstructured P2P overlay. Our goal is to provide a distributed service for Grid resource discovery that supports the features of a specialized information service such as the GT4 Index Service but is also highly scalable and fault tolerant with self-organizing and self-healing properties of P2P systems