Facebook's latest open source effort: a flash-powered database called RocksDB
Facebook has open sourced a new embedded database called RocksDB that's meant to take advantage of all the performance flash has to offer, from right on the application server. It might be a sign of best practices to come.
Context & Ripple Effects
RocksDB is the third act in Facebook's run of infrastructure giveaways: after opening its servers and data center designs in 2011 and its Corona Hadoop scheduler in 2012, the company is now handing over the storage layer itself. The backstory matters — Gigaom reported in 2011 that Facebook was trapped in what insiders called a MySQL 'fate worse than death', and this is the company's answer: skip the general-purpose relational layer for hot paths and embed a purpose-built store right on the application server.
The move also lands in a market ex-Facebookers have been circling since the MemSQL launch last year, so Facebook is now competing with its own alumni for the same 'make your database fly' audience. The pickup was unusually broad for an infrastructure release — VentureBeat, The Register, Computerworld, CIO.com and others all ran it the same day — which says more about how closely the industry watches Facebook's internals than about any single feature.
First-order effects
- Developers building latency-sensitive services get a production-hardened embedded key-value store tuned specifically for flash, removing the need to hand-tune general-purpose databases like MySQL for SSD-backed workloads.
- Facebook itself consolidates its storage stack: workloads that sat uneasily on relational infrastructure get a native path, easing the constraint behind the company's well-documented MySQL growing pains.
Second-order effects
- Database vendors and the broader MySQL ecosystem face a new competitive baseline — when a hyperscaler publishes a free flash-optimized alternative, 'good enough' relational deployments on SSDs become harder to defend on price or performance alone.
- Hardware economics shift downstream: if best practices move toward saturating flash from the application server, buyers have a reason to provision faster SSDs rather than more database servers, pressuring vendors whose margins depend on scaling out commodity nodes instead.
Third-order effects
- If hyperscalers keep releasing their internal infrastructure as open source — servers in 2011, Hadoop scheduling in 2012, storage now — their operational choices harden into de facto industry standards, narrowing how much architectural differentiation mid-tier operators can claim.
- The boundary between database product and application code blurs: embedding storage in the application server points toward a future where the database is a library choice, not a separately purchased and operated tier.
The trend: Hyperscale operators are turning their internal infrastructure into open source de facto standards, and RocksDB extends that pattern from compute and scheduling down into the storage layer itself.