What is the right term for the DynamoDB and Cassandra data model?

ScyllaDB
, ,
1

The DynamoDB Wikipedia article states that DynamoDB is a “key-value” database. However, the term “key-value” database completely overlooks a very fundamental feature of DynamoDB, that of the sort key: keys consist of two parts (partition key and sort key), and items with the same partition key can be efficiently retrieved together by the sort key.

Cassandra also has the exact same function of sorting items within a partition (which is called a “clustering key”), and Cassandra’s Wikipedia article uses the term wide colum store to describe this. While the term “wide column” is better than “key value,” it is still somewhat misleading as it describes the more general situation where an element may have a large number of unrelated columns, not necessarily an ordered list of separate items.

So my question is, is there a more appropriate term that can describe the data model of a database like DynamoDB and Cassandra: databases that, like a key-value store, can efficiently retrieve items for individual keys, but also items sorted by the key or just a part of it (the DynamoDB sort key or the Cassandra clustering key).

*Based on a question originally asked on Stack Overflow by Nadav Har’El

2

Before the introduction of CQL, Cassandra adhered more strictly to the wide column store data model, in which there were only rows identified by a row key and containing sorted key/value columns. With the advent of CQL, rows became known as partitions, and columns could optionally be grouped into logical rows via clustering keys.

Even up to Cassandra 3.0, CQL was simply an abstraction over the original thrift data model, and the concept of CQL rows within the storage engine did not exist. They were simply a sorted set of columns with a compound key consisting of the concatenated values of the clustering keys. See this article for more details. There is now native support for CQL in the storage engine, allowing CQL data models to be stored more efficiently.

However, if you think of a CQL row as a logical grouping of columns within the same partition, Cassandra could still be viewed as a wide column store. Anyway, to my knowledge, there is no other established term to describe this kind of database.

*Based on an answer on Stack Overflow by J.B. Langston