Databases are the spine of the tech industry: unsung, invisible, but critical–and beyond disastrous when they break or are deformed. This makes database people cautious. For years, only the Big Three–Oracle, IBM’s DB2, and (maybe) SQL Server–were serious options. Then the open-source alternatives–MySQL, PostgreSQL–became viable. …And then, over the last five years, things got interesting.
Some history: around the turn of this millennium, more and more people begin to recognize that formal, structured, normalized relational databases, interrogated by variants of SQL, often hindered rather than helped development. Over the following decade, a plethora of new databases bloomed, especially within Google, which had a particular need for web-scale datastore solutions: hence BigTable, Megastore and Spanner.
Meanwhile, Apache brought us Cassandra, HBase, and CouchDB; Clustrix offered a plug-and-play scalable MySQL replacement; Redis became a fundamental component of many Rails (and other) apps; and, especially, MongoDB became extremely popular among startups, despite vociferous criticism — in particular, of its write lock which prevented concurrent write operations across entire databases. This will apparently soon be much relaxed, after which there will presumably be much rejoicing. (For context: I’m a developer, and have done some work with MongoDB, and I’m not a fan.)
As interesting as these new developments–called “NoSQL databases”–were, though, only bleeding-edge startups and a tiny handful of other dreamers were really taking themseriously. Databases are beyond mission-critical, after all. If your database is deformed, you’re in real trouble. If your database doesn’t guarantee the integrity of its data and your transactions–i.e. if it doesn’t substantially support what are known as “ACID transactions“–then real database engineers don’t take it seriously:
Which is why I was so intrigued a couple of years ago when I stumbled across a booth at TechCrunch Disrupt whose slogan was “NoSQL, YesACID.” It was hosted by a company named FoundationDB, who have performed the remarkable achievement of building anACID-compliant1 key-value datastore while also providing a standard SQL access layer on top of that. Earlier this week they announced the release of FoundationDB 3.0, a remarkable twenty-five times faster than their previous version, thanks to what co-founder and COO compares to a “heart and lungs transplant” for their engine. This new engine scales up to a whopping 14.4 million writes per second.
That is a quite a feat of engineering. To quote their blog post, this isn’t just 14 million writes per second, it’s 14 million “in a fully-ordered, fully-transactional database with 100% multi-key cross-node transactions […] in the public cloud […] Said another way, FoundationDB can do 3.6 million database writes per penny.”
Impressive stuff. Impressive enough to capture the attention of enterprise database engineers, maybe. And obviously a great fit with the forthcoming Internet of Things, and the enormous amount of data that billions of connected devices will soon be constantly capturing.
But most importantly, this will push their competitors to do even better — which, in turn, will hopefully nudge the enormous numbers of enterprises still in the database Bronze Ages, running off Oracle and DB2, to consider maybe, just maybe, beginning to slowly, cautiously, carefully move into the bold new present day, in which developers are spoiled with simple key-value semantics, the full power of classic SQL queries, and distributed ACID transactions, all at the same time. In the long run that will make life better. In the interim, hats off to all the unsung database engineers out there pushing the collective envelope. You may not realize it, but they’re doing us all a huge service.