Report on Work in Progress: Massive Data Analysis Systems.

Abstract : The creation of a Massive Data Analysis System (MDAS) will enable new modes of science through improved management tools for scientific data sets, computational methods, and computational resources. To provide these capabilities, MDAS researchers are developing a software infrastructure to support data location transparency, access transparency, and conversion transparency in a heterogeneous, distributed systems. This requires a scalable software infrastructure that can manage petabytes of data, support rapid access of selected data sets, and provide support for subsequent computationally intensive analyses. The system must automate the collection of metadata describing data sets, computational methods, resources, and useers. Some of the core technologies being used to provide this functionality include object-relational database systems, archival storage systems, parallel I/O, third-party transfers, and method-level authentication. By supporting transportable methods for manipulating the data, it then becomes possible to analyze selected data sets on remote systems. The MDAS becomes an infrastructure to build next-generation operating and scheduling systems which can manage the flow of data and computation across distributed resources.