Sponsored by:

Table of Contents:

Introduction

1. Defining Availability

2. Redundancy: The Key to Carrier-Grade Availability

3. Preserving Application State: Two Techniques

4. Highly Available Database Managers

5. A Sidebar on How an RDBMS Works

6. Understanding Safeness Levels

7. Data Loss Considerations

8. MTTR Considerations

9. Application Throughput Considerations

10. Three-Way Tradeoff Summary

Self-Test

Test Answers

Glossary

PDF of this tutorial

Because a transaction is only durable after the log record has been written to disk, only a because it committed at the active database 2-Safe Durable architecture absolutely guarantees that a transaction that an application believes has been committed will be present in the standby database after it takes over from a downed active database. This guarantee is so strong that it is true even if both the active and the standby machines fail.

A 2-Safe Received architecture is reliable under any single failure. It is vulnerable only to failure of the standby before it has completely applied the copied log record. In that case, when the standby database recovers, it will see a partial transaction and will undo it.

In a 1-Safe architecture, the log record may never arrive at all at the standby database, or it may only partially arrive. This will occur even under conditions of a single failure if the active database fails before completely transmitting the log to the standby database. Under these circumstances, when the standby takes over from a downed primary, it would have no way of knowing that the transaction committed and it will undo any partial transaction activity. Unfortunately, in this case the application will believe that the transaction has committed, and unless it checks, it will not realize that data has been lost. In many systems this kind of data loss is unacceptable, and this precludes the use of a 1-Safe architecture.

Figure 12 summarizes these tradeoffs.