This is a summary of the original Cassandra white paper, available at this link.

Abstract

Casandra, in short:

• Is based on the concept of distributed storage, having no single point of failure.
• Manages a huge amount of structured data stored across cost-efficient commodity servers without sacrificing writing throughput and read efficiency.
• Doesn’t support a full relational data model.
• The data model provides a user with dynamic control over data layout and format.

Introduction:

Facebook designed Casandra to:

• Fulfill reliability and scalability requirements of the ever-growing system.
• Address storage requirements of the Inbox Search Problem. Launched in June 2008, it allowed users to search through their Facebook inbox