Though there have been substantial innovations in both file systems and storage systems over the past twenty-five years, the interface with which they communicate has remained simple and abstract. The storage stack that exists today was not developed in a coherent manner; rather, it evolved over time due to the flexible nature of this abstraction. The result is an information gap: the file system no longer understands the nature of its underlying storage, and the storage system cannot comprehend the semantics of the blocks it stores. In the end, this obscuring interface has lead to the development of independent, locally optimized, complex layers whose interactions may lead to poor performance and limited functionality.

Therefore, we believe it is time to re-examine the structure of the storage stack and the division of labor between the file system and storage system layers. Specifically, we advocate designs that enable vertical coordination and even collaboration between layers in order to achieve the goals of the entire storage stack. Furthermore, we believe the key to achieving these goals lies in information, and therefore depends on the development of informing interfaces that facilitate such vertical designs.

In this dissertation, we take three steps in this direction. First, we develop a system that automatically discovers the properties of a RAID storage system using only the logical block abstraction, transforming the obscuring interface into a basic informing interface. Second, we examine the applicability of such storage level information by designing an improved informing interface and a file system that explicitly manages the multiple disks of a storage array. Third, we develop a vertical design for collaboration between journaling file systems and RAID systems that improves the reliability and availability of the overall storage system.