Replication AI Streaming overview

DBAs want their disaster recovery system to have the most up to date database changes so that in the event of a failure, recovery can occur quicker, and retain as much data as possible. OpenEdge Replication AI Streaming is designed to meet this requirement.

In previous OpenEdge Replication implementations (without AI streaming), AI blocks are sent to the targets when they are generated, and are received by the agent and applied on the target databases. If AI data cannot be processed fast enough on the targets, it is queued on the source database machine. In case of a source crash, this data can be lost and lead to data consistency issues on the targets.

To minimize the data loss caused by AI data queued on the source, Replication AI Streaming streams the AI blocks to the targets as they are generated. The agent on the target is multi-threaded to improve performance. Replication AI Streaming allows the targets to stay as close as possible to the source database because source AI data is hardened on the targets once it is received. Replication AI Streaming is also designed to reduce the latency as AI data is less likely to be queued on the source side (PICA). The following figure demonstrates how AI Streaming works.