Storage demand in large organizations continues its inexorable increase driven by advances in applications, instruments, and sensors that generate ever more data.  Over the years, distributed storage – network-attached storage (NAS), and storage area network (SAN) – has emerged as a standard practice to provide high performance, fault resilience, and data integrity. However, depending on the quality, the vendor, discounts, and the target market, distributed storage remains expensive despite the hardware cost associated with storage falling. As a result, there has been a lot of interest in studying new methods to relieve the increasing storage demand.

In this dissertation, I have designed and implemented Storage@desk, a new low-cost virtual storage system that aggregates underutilized storage resources on distributed machines to provide a storage service with quality of service (QoS) guarantees.  As machines which Storage@desk utilizes exist in an environment that is highly unstable, Storage@desk replicates data to provide high levels of availability and reliability.  It also employs a version-based journaling algorithm for data consistency over machine failures.  Storage@desk uses encryption algorithms and the Challenge Handshake Authentication Protocol (CHAP) to provide data protection and access control. 

In Storage@desk, individual users have their own individual needs for service, that is, they may demand different levels of QoS in terms of availability, capacity, performance, etc. Each QoS property imposes various constraints and trade-offs. In this dissertation, I present QoS specification and enforcement in Storage@desk. I provide a market-based model for resource allocation, and a control-theoretical approach for performance control. I have evaluated Storage@desk by performing simulations with the real-world trace data, and benchmarks on the prototype.