Developer(s) LinkedIn Development status Active | Initial release 2009 Written in Java | |
Stable release 1.10.23 / November 10, 2016 (2016-11-10) Repository github.com/voldemort/voldemort | ||
Voldemort is a distributed data store that is designed as a key-value store used by LinkedIn for high-scalability storage. It is named after the fictional Harry Potter villain Lord Voldemort.
It is neither an object database, nor a relational database. It does not try to satisfy arbitrary relations and the ACID properties, but rather is a big, distributed, fault-tolerant, persistent hash table. A 2012 study comparing systems for storing application performance management monitoring data reported that Voldemort, Cassandra, and HBase offered linear scalability in most cases, with Voldemort having the lowest latency and Cassandra having the highest throughput.
In the parlance of Eric Brewer’s CAP theorem, Voldemort is an AP type system.
Properties
Voldemort uses in-memory caching to eliminate a separate caching tier. It has a storage layer that is possible to emulate. Voldemort reads and writes scale horizontally. The API decides data replication and placement and accommodates a wide range of application-specific strategies.
The Voldemort distributed data store supports pluggable placement strategies for distribution across data centers. Data is automatically replicated across servers. Data is partitioned meaning a single server contains only a portion of the total data. Each data node is independent to avoid central point of failure. Pluggable serialization allows rich keys and values including lists and tuples with named fields, as well as the integration with common serialisation frameworks such as Avro, Java Serialization, Protocol Buffers, and Thrift. Server failures are handled transparently. Data items are versioned, which maximizes data integrity.