Concurrent Shadow Paging: Fine-Granularity Locking with Support for Extended Lock Modes and Early Releasing of Locks

This paper describes in detail how to implement multiple concurrent transactions with fine-granularity locking in a database system using shadow paging for crash recovery. The basic idea is to store the updates separately until the transaction requests to commit, at which time the changes are patched to the global database. Commit batching is used to achieve reasonable performance; its gains are much higher with shadow paging than with log-based systems. Due to early releasing of locks, locking durations do not increase because of commit batching. 1 Introduction Traditionally, shadow paging has been considered to have poor performance and to be unsuitable for large multi-user systems [1, 3, 5, 6]. However, Kent [5, 6] has shown how to use shadow paging efficiently in a multiuser environment using page-level locking (but found it to have inferior performance compared to log-based approaches). Since then, main memory sizes have increased due to technological development, and the entire..