Availability, fairness, and performance optimization in storage virtualization systems

Storage consolidation arises due to the increasing demand for storage of massive capacity. Storage virtualization is an effective way of managing consolidated storage resources. This dissertation presents novel algorithms to resolve two critical issues in QoS guarantees in storage virtualization systems: availability guarantee and fairness guarantee. This dissertation also puts forward a unique network-centric buffer cache organization that can successfully eliminate several major overheads in data transmission within certain storage clients. The availability guarantee is supported by using replication and by translating availability requirement to conventional resource requirement. A measurement-based admission control (MBAC) algorithm is proposed that can effectively reduce the resource requirement of a virtual disk with availability guarantee to what it actually needs. The fairness guarantee is addressed in two aspects. First this dissertation proposes a new real-time disk scheduler that not only call provide normal QoS guarantees and achieve great disk utilization efficiency, but also can ensure both long-term and short term fairness among competing virtual disks. Further for the first time we address the fairness issues introduced by additional disk movement overhead for switching between virtual disks. This dissertation proposes a novel buffer cache organization to optimize the performance of storage clients, usually application servers. This cache organization call eliminate the data copying operations associated with data transportation in a special type of application servers, pass-through servers, whose main responsibility is to channel data between different external entities. Moreover, this cache organization is designed to be friendly to legacy application servers.