When Dr Michael Stonebraker speaks, the database industry listens. After all, he has been one of the leading minds in database research for around 30 years, inventing and commercialising cutting-edge prototypes and pioneering ideas that underpin many of today’s leading database management systems.
He also has a seemingly voracious appetite for start-ups, and is directly involved with a number of companies, including complex event processing firm Streambase Systems, columnar database company Vertica and his latest venture, VoltDB.
VoltDB is an open source, in-memory database designed, he says, for “extreme transaction processing”. After several years in stealth mode, the product completed beta testing earlier this year and became a commercially supported product in late May.
The database, he says, is specifically designed for companies that require “orders-of-magnitude better performance” than they can get from relational databases. Residing entirely in main memory, it is designed to automatically distribute itself over a cluster. In Transaction Processing Performance Council tests, he claims, it has outstripped conventional relational databases by a factor of 50.
In fact, says Stonebraker, companies that really care about performance no longer have any business looking at the database industry’s market leaders, or the “elephants” as he calls them.
“For 35 years, the commercial database market has followed the adage of ‘one size fits all’. The approach has been to tell customers: ‘Relational databases are the answer – now what was the question?’”
As a result, Stonebraker says, companies are dancing with elephants, using databases that are simply not designed for the kinds of high-volume transaction processing environments seen in equities trading or e-commerce websites. They’re also paying dearly in the form of hefty licensing fees, he says, which is why VoltDB is open source.
But Stonebraker has stuck to some traditions with his new venture. For example, he’s no fan of the NoSQL movement, which advocates alternatives to the common database querying language. SQL, he says, is “intergalactic data speak”, understood by everyone in the database world. Trying to propose a different approach, he says, would be like “trying to swim up the Niagara Falls”.
Another key cornerstone of VoltDB is its adherence to ACID principles – a well-established set of properties (atomicity, consistency, isolation and durability) that guarantee that database transactions are processed reliably.
“For VoltDB, it’s all about speed at scale, so that previously difficult transaction processing challenges become manageable,” Stonebraker says. “If you’re facing that kind of challenge, it’s a simple choice: you can run a conventional database on 100 nodes, with all the management burden and data centre space that that involves, or you can run VoltDB on two nodes.”
It’s a head-on challenge to the elephants but Stonebraker’s high profile in the database industry may allow VoltDB to dance with them safely.