Towards Mass Storage Systems with Object Granularity

Presented at: Eight NASA Goddard Conference on Mass Storage Systems and Technologies

ABSTRACT

Many applications, that need mass storage, manipulate data sets with KB - MB size objects. In contrast, mass storage devices work most efficiently for the storage and transfer of large files in the MB- GB range. Reflecting these device characteristics, mass storage systems typically have a file level granularity. To overcome the impedance mismatch between small objects and large files, we propose a move towards mass storage systems with object granularity. With an object granularity system, the application programmer stores and retrieves objects rather than files. The system internally maps and re-maps these objects into files. The system can adapt to changing object access patterns by remapping objects. This allows the application to be more effcient than if it were built on top of a traditional fine granularity mass storage system, employing a fixed mapping of objects. In this paper we report on investigations on the potential benefits of object granularity systems. We present an architecture that incorporates solutions to the scalability and fragmentation problems associated with object granularity.

Postscript