NoSQL 是为了解决RDBMS的3大问题：

1、可扩展性：比如跨不同机器对数据进行分区的能力。

2、提高性能：在某些情况下，RDBMS的性能比较低。

3、固定的Schema：RDBMS有很多好处（参照完整性，关系，触发器等等），但是强迫你把所有数据存储在一个固定的模式下（Schema）。

基本来说有几种不同类型的NoSQL数据库：

1、Key-Value：如Scalaris, Tokio Cabinet, Voldemort；

　　store data in key/value pairs: very efficient for performance and higly scalable, but difficult to query and to implement real world problems

2、Tabular：如Cassandra, Hbase, Hypertable, Google Bigtable；

　　store data in tabular structures, but columns may vary in time and each row may have only a subset of the columns

3、Document-Oriented：如CouchDb, MongoDb, Riak, Amazon SimpleDb；

　　like Key/Value but they let you nest more values for a key. This is a nice paradigm for programmers as it becomes easy, specially with script languages (Python, Ruby, PHP...), to implement a one to one mapping relation between the code objects and the objects (documents) in the database

4、Graph：如Neo4J, InfoGrid, AllegroGraph；

　　stores objects and relationships in nodes and edges of a graph. For situations that fit this model, like hierarchical data, this solution can be much much faster than the other ones.

MongoDb is a document oriented NoSQL database. With such a database it is very easy to map the programming objects (documents) we want to store to the database. JSON is a very viable standard to do this mapping, and so MongoDb does: it stores JSON documents in the database.

To makes performance better JSON is stored by MongoDb in a efficient binary format called BSON. BSON is a binary serialization of JSON-like documents and stands for Binary JSON.

http://www.rackspace.com/blog/nosql-ecosystem/