Parallel On-the-Fly Reading of an Entire Database Copy

Abstract

To recover from media failures, a database is 'restored' from an earlier backup copy. A recovery log of transactions is used to roll forward from the backup version to the desired time (the current time). High availability requires that the backup copying be fast, and be in parallel with on-going update activity. It also necessitates, frequently obtaining a consistent copy of an entire database. Such concurrent generation of a database copy, interferes with system activity. It introduces blocking and delays for many update transactions. We propose an algorithm that reads current database entities without interference with update activity. The algorithm is simple to implement as compared with previous proposals. It assigns a color to each entity read by the global-read. Normal transactions commit by declaring a color for the committed updates. Subsequently, these markings are used for generation of a consistent copy of the entire database.

Keywords: Concurrency Control Algorithms, Database Recovery, Global-read Transactions, Locking, Long-lived Transactions, Read-only Transactions, Synchronization, Transaction Processing