Distributed High Performance Grid Information Service Mehmet S. Aktas 1, 2 , Geoffrey C. Fox 1, 2, 3 , Marlon Pierce 1 1 Community Grids Laboratory, Indiana University 501 N. Morton Suite 224, Bloomington, IN 47404 {maktas, gcf, mpierce}@cs.indiana.edu http://www.communitygrids.iu.edu/index.html 2 Computer Science Department, School of Informatics, Indiana University 3 Physics Department, College of Arts and Sciences, Indiana University Abstract. We introduce a distributed high performance Grid Information Ser- vice Architecture, which forms a metadata replica hosting system to manage both highly-dynamic, small-scale and relatively-large, static metadata asso- ciated to Grid/Web Services. We present an empirical evaluation of the pro- posed architecture and investigate its practical usefulness. The results demon- strate that the proposed system achieves high-performance and fault-tolerance with negligible processing overheads. The results also indicate that efficient de- centralized Grid Information Service Architectures can be built by utilizing publish-subscribe based messaging schemes. 1 Introduction Information Services address the challenging problems of announcing and discover- ing resources in Grids. Existing implementations of information services present several challenges. Firstly, most of the existing approaches to Information Services have centralized components and do not address high performance and fault-tolerance issues [1, 2]. Secondly, previous solutions do not address distributed information management requirements of dynamic Grid/Web Service collections such as efficient request distribution and replica-content placement strategies [2]. Thirdly, in order to optimize performance in metadata access, information services should be able to relocate metadata to nearby locations of interested entities. Therefore, we see a great- er need for a system to facilitate resource discovery in peer-to-peer/grid service- oriented architecture based applications. In order to address these challenges, we designed, built and evaluated a Grid Information Service called Hybrid Grid Informa- tion Service (Hybrid Service) which forms a replica hosting system. This paper focus- es on the experiences gained in designing and building a replica hosting system for the Hybrid Service. The organization of the rest of the paper is as follows. Section 2 reviews the major solutions in state of art of the studies covered in this study. It gives an extensive sur- vey on the previous metadata management solutions for replica hosting systems. The previous solutions are analyzed followed by discussions on the reasons why the pre-