Netflix has completed a major consolidation of its relational database infrastructure onto Amazon Aurora, reporting performance improvements of up to 75% and a 28% reduction in costs. The streaming giant’s move from self-managed PostgreSQL on EC2 to the managed Aurora service highlights a broader industry trend toward offloading operational overhead to cloud-native database platforms.
Facing challenges with a fragmented database landscape, Netflix’s Online Data Stores (ODS) team struggled with the operational toil of managing deployed custom binaries, patching, and manual scaling. The previous architecture, relying on self-managed distributed PostgreSQL, suffered from inconsistent latency and complex failure recovery procedures. By migrating to Amazon Aurora PostgreSQL-Compatible Edition, the team aimed to unify their strategy while retaining the familiarity of the PostgreSQL engine for their developers.
The results of the migration were immediate and significant across critical microservices. Spinnaker, Netflix’s continuous delivery platform, saw a 50% reduction in average latency, dropping from 67.57ms to 41.70ms. The Policy Engine service experienced even more dramatic gains, with latency on key endpoints falling by 75%, from 26.72ms to 6.51ms. These improvements are attributed largely to Aurora’s architecture, which separates compute from storage and utilises a log-based write approach. This design allows the database to allocate 75% of instance memory to shared buffers, significantly more than the typical 25–40% in standard PostgreSQL.
Ammar Khaku, Staff Software Engineer on the Netflix Online Data Stores team, commented on the operational shift: “We no longer have to build and deploy custom binaries on EC2 with internal security and metrics-related patches. Switching to off-the-shelf managed Aurora PostgreSQL lets us focus on business logic and data access patterns.”
Netflix’s experience mirrors that of other large enterprises transitioning to managed cloud databases. Samsung Electronics migrated over 1.1 billion users from legacy Oracle databases to Aurora, citing similar drivers regarding licensing costs and the flexibility required for microservices. Panasonic Avionics also reported an 80% reduction in costs and improved query speeds after moving to the platform. These cases suggest a consistent pattern in which the elimination of licensing fees and administrative overhead drives ROI for migration.
However, engineers evaluating similar moves should note that Aurora is not a universal remedy for all workloads. While Netflix benefits from the shared storage model, independent benchmarks suggest nuances in performance. For instance, tests by Tiger Data indicated that for specific time-series heavy workloads, specialised PostgreSQL extensions like Timescale could offer faster ingest rates and lower storage costs than Aurora. Furthermore, distributed SQL alternatives such as CockroachDB or TiDB offer multi-writer capabilities, addressing Aurora’s single-writer limitation, which can become a bottleneck for write-heavy global applications.
Despite these architectural distinctions, the operational benefits for Netflix have been clear. The migration utilised Aurora’s fast failover capabilities, where read replicas can be promoted to writers in under 100 milliseconds, significantly improving system availability compared to the manual interventions previously required. This reliability, combined with the removal of undifferentiated heavy lifting, allows the streaming provider to support its massive user base with greater agility.
