Migrating Enterprise Storage Applications to the Cloud
Cloud computing has emerged as a model for hosting computing infrastructure and outsourcing management of that infrastructure. It offers the promise of simplified provisioning and management, lower costs, and access to resources that scale up and down with demand. Cloud computing has seen growing use for Web site hosting, large batch processing jobs, and similar tasks. Despite potential advantages, however, cloud computing is not much used for enterprise applications such as backup, shared file systems, and other internal systems. Many challenges need to be overcome to make cloud suitable for these applications, among them cost, performance, security, and interface mismatch.
In this dissertation, I investigate how cloud infrastructure can be used for internal services in an organization, with a focus on storage applications. I show how to design systems to address the challenges of using the cloud by building two example systems. The first, Cumulus, implements file system backup to a remote cloud storage provider. With Cumulus I consider the constraints imposed by the interface to cloud storage, and how to work within those constraints to minimize the cost. The second system, BlueSky, is a shared network file server which is backed by cloud storage. BlueSky builds on ideas from Cumulus to reduce system cost. It relies on cloud storage for data durability, but provides good performance by caching data locally. I additionally study how file system maintenance tasks can be offloaded to the cloud while protecting the confidentiality and integrity of file system data. Together, these two systems demonstrate that, despite the challenges, we can bring the benefits of the cloud to enterprise storage applications.