Best practices for HADR performance
Hao Yan, IBM
Db2 HADR (High Availability Disaster Recovery) is a Db2 LUW feature for high availability and disaster recovery. In case of fatal error on primary, the HADR can failover to standby, making it new primary to achieve HA (High Availability). Only in PEER state, the failover operation can be done without data loss. Therefore, it is vital for the HADR standby host to keep up with primary host and stay in PEER state, to allow a fast and smooth failover.
To achieve this goal, there are two major parts.
- The standby needs to quickly catch up to primary from a behind position to reach PEER state.
- The standby needs to maintain the PEER state as long as possible, even when stressed with high volumes of workloads.
Both parts are affected by HADR standby replay performance, but each part also has a few unique tricks. On the other hand, the primary performance could be negatively impacted by standby if standby is falling behind. With some configurations it’s possible to alleviate the issue. In this talk, we’ll go over the best practices for HADR performance in various scenarios and the reasons behind them.