Cassandra gets performance tuning options
The open source NoSQL and Big Data database Apache Cassandra has been updated to version 0.6.6 and now allows users to tune performance. The changes that have been made are based on real world experience with customers and users. They include the ability to adjust Cassandras's indexing interval to make it more memory efficient with large amounts of small rows with "cold data" and the ability to control when the JVM should trigger garbage collection to avoid the database being paused for several seconds.
Other changes include the option to increase throughput by disabling "read repair", the process by which Cassandra makes replicas consistent, improvements to cache handling and the removal of deleted data. A full list of the changes made in 0.6.6 is available in the change log. Jonathan Ellis, Cassandra project chair and co-founder of Riptano, a Cassandra-centric support and services company, explains more about the changes in a blog posting.
Originally developed by Facebook, Cassandra was submitted to the Apache Software Foundation's Incubator in 2009. In February, it was promoted from the incubator to become an Apache top level project and Cassandra 0.6 was released in March. Cassandra offers a high availability, decentralised and scalable data store which has been used for sites such as Digg, Reddit, Twitter, Cisco's WebEx, CloudKick and Rackspace. Cassandra implements an "eventually consistent" model which trades-off consistency of data stores in the system for availability.
Cassandra 0.6.6 is available to download and is licensed under the Apache Licence 2.0.