MongoDB's feline brains trust sink their claws into NoSQL

Relational payoff in the wings?

By any measure, BerkeleyDB was a hit. It became the world’s most widely deployed embedded and open-source database, meaning that the company which did the most work to maintain it, Sleepycat, got swallowed by Larry Ellison’s database giant Oracle in 2006.

Mike Olson was Sleepycat’s business chief. He’s now CEO of Hadoop venture Cloudera, which is talking about an IPO.

Michael Cahill and Keith Bostic – two of the brains behind both Berkeley DB and Sleepycat – followed up with database storage engine specialist WiredTiger.

Cahill and Bostic were bought with WiredTiger in December 2014 by MongoDB, while last week saw the release of MongoDB 3.0 - the first fruits of the union. MongoDB is betting Cahill, Bostic, their team and WiredTiger will deliver great things for their open-source document-oriented store.

“We can access an incredible amount of experience in this area - collectively they have more than 75 years of innovation in this space. There aren’t many people who possess their level of experience,” Kelly Stirman, MongoDB director of products, told The Reg.

“We can continue to develop and work with that technology and gain better visibility and control over how WiredTiger develops in the future,” he added.

MongoDB 3.0 sees a fundamental re-write of the MongoDB core storage engine and back-end architecture, based on WiredTiger. WiredTiger makes MongoDB a more flexible and scalable NoSQL database.

MongoDB had been a slave to locking – make a change to one field or object and the entire database was locked while changes were replicated. In the understated lexicon that is dev-speak, this was, er, “sub-optimal.”

WiredTiger brought in document-level locking, which meant greater flexibility in environments where lots of updates and changes are made. That means places with lots of machine data flooding in – for example, IoT apps, website updates or data analysis.

MongoDB can now handle lots of write-heavy loads out of the box and claims a performance improvement on write throughput of between seven and 10 times on the previous version.

WiredTiger also adds compression to Mongo DB. The NoSQL document store will now eat between 50 to 80 per cent less disk space than before.

The biggest impact of WiredTiger is it sees MongoDB embark upon a policy of working with lots of different engines and data types. MongoDB 3.0 has three storage engines: the original MMAP, WiredTiger plus in-memory for those shy of writing their data to disk.

MongoDB lets you work across different data types, too – it’s not either/or – while keeping the same JSON programming model.

Other data heads are excited by the changes bought with WiredTiger in MongoDB 3.0.

“For me, it’s a new technology in the database space – WiredTiger is a big deal,” the chief executive of Object Rocket, Chris Lalonde, told The Reg.

“It’s a really big deal for them, a big deal in the NoSQL space because nobody lets you change the back end. The core characteristics of the database mean you can have one Mongo cluster that’s very good a reads and change the back end module to make MongoDB very, very good at writes. In places where other databases have dished on Mongo they have a good answer with WiredTiger.”

ObjectRocket is the Rackspace-owned company that delivers MongoDB and Redis as services.

One storage model is particularly intriguing, especially to the regular enterprise user – rows and columns, a signature of relational databases. WiredTiger supports row and columnar-oriented storage – orthogonal, to the world of NoSQL.

“There’s enormous opportunity left for the storage engine – we are just beginning to explore the capabilities in the technology,” Stirman said.

“For high performance scale and classification, there’s no reason we couldn’t expand from a document models to table-based models or to graph. The reality is no-one data model works for all applications,” he added.

Should we expect MongoDB to fire up relational for broader enterprise uptake, to copy the success of Sleepycat in its particular market?

“We’re not committed to exposing that in the database in a specific release but you have to consider, because it’s in the storage engine already, [that] that’s a pretty significant step in that direction, should we choose to pursue it,” Stirman told The Reg. ®

Similar topics


Send us news

Other stories you might like