Database Storage Management in a Shared Virtual Memory Environment

Shared Virtual Memory (SVM) provides a shared memory abstraction in systems that do not have a physically shared memory. Several architectures that provide this sharing abstraction without the scalability limitations of physical sharing have been proposed. However, there has not been much work in the area of database systems to exploit this scalable abstraction. In earlier work we proposed a scalable sharing database architecture that exploits the SVM abstraction for better scalability and resource utilization. In this paper we propose a database storage management technique that exploits the SVM abstraction to better utilize the global memory available. The technique uses remote memory as a part of the overall storage hierarchy. Through this technique the database system reduces the overall amount of disk I/O, thus improving performance by about 20%. Furthermore, we also discuss how this storage management technique is implemented in existing systems and how it can improve even uniprocessor database performance.