At the recent Microsoft Ignite conference, a significant announcement was made regarding the early preview of Azure HorizonDB, a managed database service designed to be compatible with Postgres, specifically tailored for enterprise workloads.
Key Features and Performance Enhancements
Azure HorizonDB is engineered to support an impressive scale of up to 3072 vCores across both primary and replica nodes. Its auto-scaling shared storage capability accommodates databases as large as 128TB, boasting sub-millisecond multi-zone commit latencies. According to Affan Dar, VP of Engineering at Microsoft, and Charles Feddersen, partner director at Microsoft:
Developers benefit from the robust Postgres ecosystem and seamless integration with Azure’s advanced AI capabilities, while enterprises can gain a secure, highly available, and performant cloud database to host their business applications.
A notable addition is the AI Model Management feature, which integrates Microsoft Foundry models directly into the customer database. This feature claims to enhance vector search speeds by up to three times, utilizing advanced DiskANN filtering as opposed to traditional HNSW methods. The development team elaborates on this innovation:
Advanced Filtering addresses a common problem in vector search – combining vector search with filtering. DiskANN Advanced Filtering solves this by merging filter and search into a single operation. As the graph of vectors is traversed during the vector search, each vector is simultaneously checked for filter predicate match, ensuring that only the correct vectors are retrieved.
Innovative Storage Architecture
Adam Prout, partner architect at Microsoft and former CTO and co-founder at SingleStore, emphasizes the advantages of HorizonDB’s architecture:
HorizonDB pairs upstream compatibility with disaggregated storage designed to make Postgres really shine. We’re pushing as much replication and durability work as we can into the storage layer, leaving more CPU for PostgreSQL to run queries and transactions. This approach reduces commit latencies, accelerates failovers, and scales reads across replicas while integrating with Azure security and AI tooling.
To further streamline database management, Microsoft is enhancing the PostgreSQL extension for Visual Studio Code, which has now reached general availability.
Comparison with Existing Offerings
Azure HorizonDB complements Microsoft’s existing managed PostgreSQL solutions, including Azure Database for PostgreSQL, a general-purpose managed database for traditional workloads, and Cosmos DB for PostgreSQL, which is based on Citus for horizontally scaled workloads. What sets HorizonDB apart is its shared-storage, scale-out compute architecture, specifically optimized for AI workloads.
In a competitive landscape, Microsoft faces challenges from other cloud providers like Amazon Aurora and Google’s AlloyDB. Luke Fangman, managing director of cloud and AI at Microsoft, notes:
For those in the know, we have been waiting for Microsoft to finally take on Aurora to give enterprise true choice when you need globally scalable & performant PostgreSQL engines.
For those interested in a deeper understanding, the recording of the Ignite session titled “Azure HorizonDB: Deep Dive into a New Enterprise-Scale PostgreSQL” is now accessible. During the early preview, HorizonDB will be available in a limited selection of regions.
About the Author
Renato Losio
Show moreShow less
