There are two primary issues facing database systems designed for on-line transaction processing (OLTP): scalability and performance. Traditional disk-based OLTP architectures are the result of a now decades-old hardware assumption that main memory can only store a small subset of the data at any given time. In order address the issue of high disk latency, such architectures employ concurrent execution strategies with complex and expensive mechanisms to ensure correctness. Such an architecture addresses the scalability issue by incorporating high-capacity persistent storage, but fails to meet the performance demands of modern OLTP applications. More recently, architectures have been proposed that take advantage of the huge advances in main memory capacity. By designing an OLTP architecture under the assumption that all data would fit in memory, these main memory database architectures offer significant performance advantages over disk-based architectures. However, they are bounded by the capacity of main memory, and make no use of the significantly higher capacity of persistent storage. In this dissertation, we present a novel architecture called anti-caching that addresses both the performance and scaling needs of modern OLTP applications. Anti-caching works as an extension to a main memory architecture by storing frequently accessed data in main memory and moving cold data to the anti-cache on persistent storage. We present several lightweight mechanisms for tracking data accesses and for efficient and asynchronous movement of data. Finally, we consider the use of next-generation hardware in the form of non-volatile memory, and explore its use in existing OLTP architectures. Taken together, the contributions in this dissertation detail an OLTP architecture well-suited for current and future hardware assumptions that addresses both the performance and scalability of modern OLTP applications.
DeBrabant, Justin A.,
"Anti-Caching and the Use of Non-Volatile Memory in OLTP Databases"
(2015).
Computer Science Theses and Dissertations.
Brown Digital Repository. Brown University Library.
https://doi.org/10.7301/Z0SQ8XSZ
