The third chapter of the book is divided into two main sections. The first one explains the most typical data structures to store information in a database and the second describes the differences between OLTP and data warehouse databases. This post will not deep into much detail so I would recommend you to read the chapter to fully understand how these two data structured models work in details.
Storage and Retrieval
Log-structured and B-Trees are the most common data structures used.
The log-structured model allows appending data to files and deleting obsolete files, but never updates an existing file that has already been written. Some examples of log-structured databases are Bitcask, LevelDB, Cassandra, HBase and Lucene.
On the other hand, we have the B-tree type model where the storage is divided into fixed pages that can be overwritten. The majority of the relational databases use this model.
As an unwritten rule, log-structured models are faster for writes and B-trees are better on readings. Reads are slower on log-structured because they have to check several data structures at different stages of compaction.
Of course, you will need to test which model suits you better for each specific case.
OLTP and data warehouse databases
OLTP (Online transaction processing) systems are typically user-oriented, which means they have to deal with a high volume of requests. They are designed to be effective when only touching a small number of records in each query.
On the other side, data warehouse databases are analytic systems prepared to be used by analytics people. They handle a much lower volume of requests but these queries are usually very demanding, scanning millions of records in a short time.