Traditional database systems rely upon a proven set of tools to guarantee ACID properties without compromising performance: a buffer manager to mediate the transfer of data between fast in-memory processing and slow disk-based persistent storage, latching and locking to coordinate concurrent access to data, and logging to enable the recovery, verification, and repair of committed data. These tools are built on code bases that are 10-30 years old and designed for hardware assumptions nearly the same age. Modern hardware technologies such as fast persistent memories and multicore break those assumptions, turning the traditional proven tools into the new bottlenecks. Our goal is to rethink the traditional tools so that they will not be bottlenecks. Here, we review some of the concurrency-related bottlenecks that face the modern transactional storage management system and survey state of the art techniques that allow these traditional tools to provide intended functionality without becoming bottlenecks. © 2013 Springer-Verlag Berlin Heidelberg.
CITATION STYLE
Kuno, H., Graefe, G., & Kimura, H. (2013). Making Transaction Execution the Bottleneck (Instead of All the Overheads). In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 7813 LNCS, pp. 71–85). Springer Verlag. https://doi.org/10.1007/978-3-642-37134-9_6
Mendeley helps you to discover research relevant for your work.