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

In this paper we describe Cumulus, a system for efficiently implementing filesystem system backups over the Internet. Cumulus is specifically designed under a thin cloud assumption--that the remote datacenter storing the backups does not provide any special backup services, but only provides a least-common-denominator storage interface (i.e., get and put of complete files). While Cumulus can thus use virtually any storage service, we show that it still provides efficiencies comparable to integrated approaches.

Pre-2018 CSE ID: CS2008-0927

Due to the lack of deployment of a network-layer multicast service, many overlay multicast protocols have been designed and deployed across the Internet to support content distribution. To our knowledge, however, none have provided a rigorous analysis of the problem or the effectiveness of their proposed solutions. Here, we set aside the engineering challenges of protocol design to focus on the fundamental graph problem. We begin by formulating the Overlay Network Content Distribution (OCD) problem and show that variants that attempt to optimize for either speed or bandwidth utilization are NP-complete. Using both a time-indexed Integer Program and a branch-and-bound search strategy, we calculate optimal solutions for small graphs. While solutions to OCD provide performance bounds, realistic deployment scenarios will not have global network information. Hence, we introduce an on-line variant, the Local-knowledge Overlay Content Distribution (LOCD) problem and show that no constant-competitive approximation exists. Instead, we present several heuristics that perform well in realistic topologies. We conclude with an evaluation of our global and local heuristics and enumerate a number of open problems.

Pre-2018 CSE ID: CS2005-0824

Modern organizations face increasingly complex information management requirements. A combination of commercial needs, legal liability and regulatory imperatives has created a patch-work of mandated policies. Among these, personally identifying customer records must be carefully access-controlled, sensitive files must be encrypted on mobile computers to guard against physical theft and intellectual property must be protected from both exposure and ``poisoning.'' However, enforcing such policies can be quite difficult in practice since users routinely share data over networks and derive new files from these inputs -- incidentally laundering any policy restrictions. In this paper, we describe a VMM system called Neon that transparently labels derived data using byte-level ``tints'' and tracks these labels end-to-end across commodity applications, operating systems and networks. We demonstrate that this mechanism allows the enforcement of a variety of data management policies, including data-dependent confinement, intellectual property management, and mandatory I/O encryption.

Pre-2018 CSE ID: CS2008-0934