I am working on publishing the English version of “PostgreSQL 9.6 Performance Story” on Amazon. Below is a preview of Chapter 2.
Shared buffer is an essential component for efficient IO processing.
Because of this importance, the DBMS buffer manager will be highly optimized. In other words, if you set the Shared Buffer to a reasonable size (several Gb or tens of Gb, or even hundreds of Gb), there may not be a performance problem caused by Shared Buffer.
Then why? Do I need to know how Shared Buffer works? It is doubtful.
This is also the question I had when I started the Shared Buffer study.
However, as the research on Shared Buffer was repeated, it was found that Shared Buffer has characteristics. IO strategy, ring buffer, and clock sweep algorithm. Without knowing these characteristics accurately, it can be difficult to determine the cause of sudden slowing of IO processing.
So let’s get to the point.