History[03][06]

VoltDB originated from H-Store, a research project from Brown U and MIT under Michael Stonebraker, Samuel Madden, Andy Pavlo, etc. In 2009, it started as a spinoff in Boston, MA.

Checkpoints[07]

The snapshot creation could either block all the other transactions or be done in a non-blocking way (e.g. Copy-On-Write).

Concurrency Control[04][05][08]

VoltDB schedules transactions deterministically on each partition of the data in the cluster. For transactions that span across partitions, a global serial order controller is responsible for determining the execution order. MVCC mode is only used when taking snapshots.

Data Model[09]

VoltDB is based on the standard relational model.

Foreign Keys[10][11]

FOREIGN KEY and CHECK are not currently supported.

Indexes[12]

Apart from the default tree index, VoltDB also supports hash index.

Isolation Levels[04][05][08]

VoltDB supports serializable isolation level with ACID guarantee. The deterministic ordering of transactions is achieved by controllers at each data partition.

Joins[13][14]

Table joins are only supported if the result is smaller than 50MB. The join ordering must be specified if the query joins more than 6 tables.

Logging[15]

VoltDB supports transaction-level logging for all the stored procedures that are invoked. By command logging, the amount of memory overhead is minimized since a single stored procedure invocation could contain arbitrary number of SQL statements. Only logs after the latest snapshot are kept, and the logs are flushed to the disk at a regular interval. This feature is only available in the enterprise edition.

Query Compilation[16][04]

Query Execution

Query Interface[16][09][17]

Apart from supporting most of the standard SQL statements, VoltDB also allows the user to define pre-compiled stored procedures.

Storage Architecture[07]

VoltDB is designed to be an in-memory relational DBMS. Its data persistence is achieved by snapshots stored on disk.

Storage Model[18]

VoltDB is only designed for fast OLTP queries.

Stored Procedures[16][04]

The stored procedure in VoltDB allows the user to define a transaction once and to be invoked later with given parameters. Stored procedures are pre-compiled to reduce execution time. Temporary queries are automatically comipled and executed as a stored procedure. VoltDB supports stored procedures written in SQL or Java.

System Architecture[19][20]

VoltDB runs on a sharded DBMS cluster. Data partitioning and replication is done automatically across the cluster

Views[21]

Views in VoltDB are materialized to improve read speed, and all updates to the original table will have a performance impact.