OpenAI Updates Habitat Storage Platform for 1 Billion ChatGPT Users
OpenAI has restructured its Habitat storage platform, which manages 500 petabytes of data, to serve over 1 billion users. The company resolved scalability issues by transitioning to a multi-tenant architecture on Azure Cosmos DB.
OpenAI has completed the second phase of scaling its internal storage platform, called Habitat, which serves over 1 billion users across 40 different geographic regions. Managing more than 500 petabytes of data, the system underwent an architectural transformation to overcome performance and reliability challenges.
Initially utilizing a multi-tenant Azure Cosmos DB structure, OpenAI encountered performance bottlenecks due to "noisy neighbor" issues and logical partition key limits. This situation necessitated a shift to single-tenant containers for critical services, but this method introduced operational burdens such as capacity planning complexity and growth instability.
OpenAI collaborated with the Azure team to develop a new multi-tenant model. This model features hot key throttling, request tiering for quality of service, and improved serverless mode capabilities that allow physical partitions to scale up to 10,000 RU/s. The return to a shared container model eliminated resource provisioning issues while successfully maintaining workload isolation.
Sources
- openai.com · Link