VoltDB this week rolled out an enhanced version of its flagship database, one of a new class of purpose-built, high-velocity databases that bear the moniker "new SQL." Version 3.0
As a "new SQL" database, VoltDB is meant to reap the benefits of semiconductor memory while avoiding some drawbacks of the disk drive memory that was so central to the rise of relational database management systems (RDBMSs). Moreover, unlike a pack of newly released NoSQL databases, VoltDB and other new SQL offerings natively support SQL-style programming.
"VoltDB is an in-memory relational database aimed at rapid, high-volume data processing and ingestion," said Matthew Aslett, research manager of data management and analytics at London-based 451 Research Group, referring to the increased need of many applications to "ingest" data, or quickly populate a database with active data feeds.
For more on New SQL
Find out how Hadoop will coexist with relational data in 2013
Review Stonebraker's outlook circa 2012
Learn how an effective big data strategy helped an advertising firm attract clients
See coverage of last year's Usenix event
Aslett said the company has seen early successes in applications in the financial services market, but that its software could find use in a wide variety of Web-oriented applications. "Its sweet spot would be transactional Web applications," he asserted.
VoltDB's technology derives from initial academic work on an H-Store transaction processing engine undertaken by database stalwart Michael Stonebraker and others. H-Store was commercialized as VoltDB, which saw its first release in the form of an open source community edition version 1.0 in 2010.
As an in-memory SQL RDBMS, it is designed to provide ultrahigh throughput. While supporting SQL and ACID transactions, VoltDB largely forgoes disk-related RDBMS locking, latching and buffer management steps that create high-latency and processing overhead for transactions, according to Stonebraker, who serves as VoltDB CTO.
In a lengthy career, Stonebraker had a major hand in such database undertakings as Ingres, Postgre SQL, and such recent noted database startups as Streambase and Vertica, which was later acquired by Hewlett-Packard. This week, at the hack/reduce developer space near the Massachusetts Institute of Technology in Cambridge, Mass., he told an audience that generalized databases will find competition from specialized types in years to come.
That could mean a more varied database landscape than in the past. "I think five to ten architectures will survive," Stonebraker said. He calls the specialized databases "purpose-based," and said some of these designs have an edge when it comes to faster performance.
When it comes to speed, "purpose-based systems are bound to win," he said. The VoltDB, he emphasized, gains performance from its use of main memory, stored procedures and deterministic scheduling. Still, experience in the field has led VoltDB to make additional changes for emerging applications that must very quickly ingest great helpings of Web data.
Ryan Betts, a software engineer at VoltDB, said the company has created a very fast export-and-extract system, and has eased the programmer's tasks for integrating with analytical stores such as those from data warehouse appliance vendors Netezza, which is owned by IBM, and Vertica.
"Our 'use case' is almost always around high velocity data ingestion," said Bruce Reading, president and CEO at VoltDB.