http://perspectives.mvdirona.com/2008/07/12/FacebookReleasesCassandraAsOpenSource.aspx
- Designed to handle very large amounts of data
- Spread out across many commodity servers
- Provides highly available service with no single point of failure.
- Initially developed by Facebook and powered Inbox Search until late 2010
- Jeff Hammerbacher: BigTable data model running on an Amazon Dynamo-like infrastructure (distributed storage system)
- write ahead logging and indexing
- Structured key-value store
- Tunable consistency
- Keys map to multiple values, which are grouped into column families.
- Column families are fixed when database created
- Columns can be added to a family at any time.
- Hybrid between column-oriented DBMS and row-oriented store.[
- BigTable modeling
- eventual consistency
- Gossip protocol for cluster membership maintenance
- master-master serving read/write requests inspired by Dynamo
- Writes:
- Write to arbitrary node in Cassandra cluster
- Request sent to node owning the data
- Node writes to log first and then applied to in-memory copy
- No locks in critical path
- Sequential disk accesses
- Behaves like a write through cache
- Atomicity guarantee for a key
- Always writable
