Cassandra is a data store that is designed to replicate aspects of Bigtable’s query interface and scaling, while operating directly on consumer hardware. To do this, it draws on leaderless replication strategies from Dynamo.
Cassandra lacks Bigtable’s versioning. However, it introduces a SQL-like query language, CQL, which allows selection of multiple columns at a time.
Cassandra’s storage architecture is similar to DynamoDB, but using SSTables. It operates as a leaderless collection of nodes. Each row is assigned to a “coordinator” node, which in turn replicates to some number of other nodes. Writes are always accepted, being reconciled using vector clocks.
Unsurprisingly, Cassandra is extremely available and scalable at the expense of consistency.