Improving Data Management and Data Movement Efficiency in Hybrid Storage Systems

University of Minnesota Ph.D. dissertation.July 2017. Major: Computer Science. Advisor: David Du. 1 computer file (PDF); ix, 116 pages.In the big data era, large volumes of data being continuously generated drive the emergence of high performance large capacity storage systems. To reduce the total cost of ownership, storage systems are built in a more composite way with many different types of emerging storage technologies/devices including Storage Class Memory (SCM), Solid State Drives (SSD), Shingle Magnetic Recording (SMR), Hard Disk Drives (HDD), and even across off-premise cloud storage. To make better utilization of each type of storage, industries have provided multi-tier storage through dynamically placing hot data in the faster tiers and cold data in the slower tiers. Data movement happens between devices on one single device and as well as between devices connected via various networks. Toward improving data management and data movement efficiency in such hybrid storage systems, this work makes the following contributions: To bridge the giant semantic gap between applications and modern storage systems, passing a piece of tiny and useful information (I/O access hints) from upper layers to the block storage layer may greatly improve application performance or ease data management in heterogeneous storage systems. We present and develop a generic and flexible framework, called HintStor, to execute and evaluate various I/O access hints on heterogeneous storage systems with minor modifications to the kernel and applications. The design of HintStor contains a new application/user level interface, a file system plugin and a block storage data manager. With HintStor, storage systems composed of various storage devices can perform pre-devised data placement, space reallocation and data migration polices assisted by the added access hints. Each storage device/technology has its own unique price-performance tradeoffs and idiosyncrasies with respect to workload characteristics they prefer to support. To explore the internal access patterns and thus efficiently place data on storage systems with fully connected (i.e., data can move from one device to any other device instead of moving tier by tier) differential pools (each pool consists of storage devices of a particular type), we propose a chunk-level storage-aware workload analyzer framework, simplified as ChewAnalyzer. With ChewAnalzyer, the storage manager can adequately distribute and move the data chunks across different storage pools. To reduce the duplicate content transferred between local storage devices and devices in remote data centers, an inline Network Redundancy Elimination (NRE) process with Content-Defined Chunking (CDC) policy can obtain a higher Redundancy Elimination (RE) ratio but may suffer from a considerably higher computational requirement than fixed-size chunking. We build an inline NRE appliance which incorporates an improved FPGA based scheme to speed up CDC processing. To efficiently utilize the hardware resources, the whole NRE process is handled by a Virtualized NRE (VNRE) controller. The uniqueness of this VNRE that we developed lies in its ability to exploit the redundancy patterns of different TCP flows and customize the chunking process to achieve a higher RE ratio