Log organization structure clustered based on transaction aggregation and method for realizing corresponding recovery protocol thereof

Abstract

The invention discloses a log organization structure clustered according to transaction aggregation and a recovery protocol based on the log organization structure clustered, which can be applied to a transactional data management system of a large-sized computer. A log file is sequentially organized to a plurality of log fragments and each log fragment is used for storing the log content of the same transaction and reserving a transaction number as well as a preceding log fragment pointer of the transaction; a data page number involved in a log entry of the same fragment is stored in the form of an array. When the system is operating, each transaction only writes its own log fragment and writes the log fragment in the log file when the transaction is submitted. In a recovering process, the system can be recovered to a lasting and consistent state by scanning all the log fragments for remake and returning the log fragments of all the active transactions for return. The problem of producing bottlenecks during writing logs in the traditional transactional data management system is resolved and log amount of the system can be effectively reduced.

Description

technical field [0001] The design of the invention belongs to the field of data information processing, and in particular relates to a transactional data management system that uses log technology to ensure data consistency. Background technique [0002] In the Fung computer architecture, persistent data is stored on slower, larger external storage devices (such as hard disks), and when data is accessed, it is read into faster, smaller external storage devices. In a memory device (such as RAM) and accessed by the CPU. However, due to performance requirements, when data is updated, only its memory image is updated, and the update of external memory is delayed by the buffer management unit. [0003] In a transactional data management system, a single transaction—that is, an operation process composed of several updates—is regarded as a basic unit of work, which has two important properties. One is consistency, that is, all pairs of data The updates must be seen at the same t...

AI Technical Summary

Problems solved by technology

First of all, the volume of a single log is relatively large. Since each log must record the log sequence number S, transaction number A, and the previous item Sprev of the transaction log, the combined length of these fields often exceeds 20 bytes. When it is short, the space occupied by these domains is often close to 1 / 3 or even 1 / 2 of the entire log file space; secondly, because the action of generating logs must be globally synchronized to ensure the unique incrementality of the log sequence, and the current large computer often With dozens or even hundreds of CPUs, when updates occur intensively, it will cause scalability bottlenecks

Images