Two days ago, I saw an article About the technical documentation "All About Oracle's In-Memory Undo" on the relevant Internet. This article mainly discusses Oracle In-Memory Undo (IMU, we have talked about it in the past few years. Most of them are based on speculation. This is the first detail.
Oracle has applied for a patent for the IMU feature launched at 10 Gb ). As one of the most important components, the efficiency of Undo is directly related to the capabilities of the entire database. The old Block-based Segment (Segment refers to the concept of the storage layer) Management Mode of Undo. Changes of UNDO itself must be recorded in the Redo Log Buffer, while IMU avoids this operation (because it is memory rather than storage), thus reducing the amount of Redo generated.
In addition, because the read consistency overhead is directly stored in the memory instead of dependent on the storage segment, the overall consumption is greatly reduced, and the CPU load is also effectively reduced. Its application model should be said to be an OLTP suitable for consistent read with a large demand.
Oracle 10g uses IMU by default. The implicit parameter _ in_memory_undo can be used to disable this feature. Because it is an implicit parameter, it also reflects that this feature is not so mature. Searching Metalink involves a lot of IMU bugs. Once UNOD bugs are encountered, it may be difficult to solve non-stop DB bugs. Therefore, for systems with high availability requirements, we still need to think twice before using this feature.
The above content is a discussion of Oracle In-Memory Undo. I hope you will gain some benefits.
Article by: http://www.programbbs.com/doc/class10-2.htm