We report on a nine-month prototyping project concerning storage management in the CMS offline system. The work focused on the issue of using large disk farms efficiently. We discuss various hard disk performance characteristics which are important for physics analysis applications. It is shown that the layout of physics data on disk (clustering) has a significant impact on performance. We develop a storage management architecture which ensures high disk performance under a typical physics analysis workload. Submitted to the Eindhoven University of Technology as a diploma thesis in the post-graduate programme on software design. Keywords: CMS, prototyping, offline software, physics analysis software, ODBMS, storage management, hard disks, clustering, reclustering. This document was submitted to the Eindhoven University of Technology as a diploma thesis in the post-graduate programme on software design. 1 Summary At CERN, the European laboratory for particle physics, the fun...