An increasing number of database applications demands high availability combined with online scalability and soft real-time transaction response. This means that scaling must be done online and non-blocking. This paper extends the primary/hot standby approach to high availability with online scaling operations. The challenges are to do this without degrading the response time and throughput of transactions and to support high availability throughout the scaling period. We measure the impact of online scaling on response time and throughput using different scheduling schemes. We also show some of the recovery problems that appear in this setting.
展开▼