NoSQL databases can be classified

into four categories: Key-Value Store This is the most basic data model, where the data is stored as key-value pairs.

Columnar Store：Similar to key-value store, data is stored as key-value pairs. However,  each entry (referenced by a key) contains one or more key-value pairs instead of a value. A columnar store is essentially a two dimensional array (Tezer, 2014a). Document-Oriented Database The fundamental concept is the document, which is a group of data with internal structure that can be defined arbitrarily. A document is referenced by a key and this allows documents to nest inside one another. Unlike in relational databases, the internal structure of a document does not need to be predefined and can be changed anytime. While this makes storing data with complex or even ever-changing structure much easier, it also poses a challenge for querying as the complexity grows.

Graph Database：This uses graph structures to store data and thus, is most suitable for heavily linked data that can be best represented in graphs。

见：http://edoc.sub.uni-hamburg.de/haw/volltexte/2016/3578/pdf/Thesis.pdf 里面介绍了如何去实现一个高性能的kv数据库，嵌入式的，参考了levelDB和LMDB等。