Premium
Evaluation of the basic remote backup and replication methods for high availability databases
Author(s) -
Frank Lars
Publication year - 1999
Publication title -
software: practice and experience
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.437
H-Index - 70
eISSN - 1097-024X
pISSN - 0038-0644
DOI - 10.1002/(sici)1097-024x(19991225)29:15<1339::aid-spe284>3.0.co;2-4
Subject(s) - backup , replication (statistics) , computer science , database , backup software , commit , data loss , medicine , virology
In this paper, we describe the properties of two 0‐safe backup/replication methods and compare them with the traditional 1‐safe and 2‐safe backup/replication methods. The 1‐safe and 2‐safe backup/replication methods have been analyzed for many years, and their problems are well known. The properties of the 0‐safe designs have not yet been analyzed. The basic 2‐safe, 1‐safe and 0‐safe backup designs all use two copies of the database. However, under normal operation the database copies of the 2‐safe backup design are both consistent and up‐to‐date, whereas the 1‐safe backup design only uses one up‐to‐date copy of the database. In the 0‐safe backup design none of the databases are consistent and up‐to‐date. Therefore, this backup design is named 0‐safe . However, in the 0‐safe design the two inconsistent copies converge towards the same consistent state. In a 0‐safe design transactions update and commit the nearest database copy first and, therefore, the database copies are normally inconsistent. The inconsistency between the 0‐safe database copies must be managed by using countermeasures, which will reduce the problems of the missing distributed isolation property. We will also describe and evaluate a mixture of the 0‐safe and the 1‐safe designs, which in certain situations may have more interesting properties than the original designs. The backup/replication designs will among other things be compared with respect to their disaster recoverability, availability, performance, implementation problems and ACID properties. The largest bank in Denmark, Den Danske Bank, has converted all its important database systems from using the 1‐safe design to using the 0‐safe design. We have been cooperating with one of the major ERP (Enterprise Resource Planning) software companies in designing a distributed version of their financial management and accounting system by using the techniques described in this paper. We have now started to implement a prototype of the system. Copyright © 1999 John Wiley & Sons, Ltd.