Rethinking Cloud Storage System Software under Multi-Tenancy

Virtualization-based cloud computing has dominated today\u27s data centers by supporting consolidated servers, converged infrastructures, and horizontal scalability. To provide high resource density and low cost of ownership, current clouds typically involve a multi-tenancy architecture, which consists of a mix of layered software and hardware to realize virtually dedicated computing and storage capabilities for cloud tenants. Despite significant benefits of multi-tenancy, new challenges arise in the aspects of efficiency, fairness and customizability of cloud resource sharing. In this dissertation, we present an integrated cloud storage system consisting of three key components, StorM, vFair and BASS, to address key challenges in multi-tenancy cloud storage systems. In particular, BASS bridges the I/O data addressability gap between the storage and network stacks of cloud block storage with a novel byte-addressable storage stack. BASS not only avails the benefits of using variable-length I/O requests which avoids unnecessary data transfer, but also enables a highly efficient non-blocking approach which eliminates the blocking time of write processes. vFair addresses the fair I/O resource allocation and scheduling problem using a new I/O scheduling framework, which takes per-IO cost into consideration along with a two-level scheduling architecture. vFair strikes a good balance between fairness and I/O resource utilization, regardless of I/O workloads and patterns. Finally, StorM enables a storage middle-box platform to support software-defined cloud storage services. With StorM, tenant-specific storage data security and reliability services can be freely defined by tenants, while the infrastructure-level support of middle-boxes is transparently taken care by cloud providers