Microsoft Fabric promises transactions, analytics on one database service

Windows giant also improving Iceberg support after throwing lot in with Delta Parquet

Ignite Microsoft is set to launch a database service that can manage transactional and analytical workloads in the same system.

The software and cloud giant also promises greater support for the Iceberg table format preferred by rivals, increasing the interoperability with the native Delta in its Fabric data platform.

Last year, Microsoft launched Fabric, a new data platform that strove to bring together different data systems and services already existing in its Azure cloud, along with some new features. Notable among them was mirroring, which the vendor said replicates a snapshot of the external database to OneLake in Delta Parquet tables and keeps the replica synced in "near real time."

At Microsoft's Ignite conference this week, the vendor launched an approach to bringing together transactional and analytical workloads in a single system within Fabric.

In a blog post, Microsoft previewed Fabric Databases, which it said represented "a new class of cloud databases that bring together transactional and analytical workloads, creating a truly unified data platform."

It said: "Developers can streamline application development with simple, autonomous, and optimized for AI databases that provision in seconds and are secured by default with features like cloud authentication and database encryption."

Microsoft said features such as vector search, RAG support, and Azure AI integration would simplify AI application development. Meanwhile, user data would be "instantly available" in OneLake, Microsoft's lakehouse product, which brings together data warehouse and data lake-type workloads.

The combined analytics/transactional service is set to be first available in Fabric as a service based on the SQL Server engine, related to Microsoft's popular database product.

Arun Ulagaratchagan, corporate vice president Azure Data, told The Register that the database service would "combine operational data with analytics and AI into a converged data platform that unlocks new scenarios for insight-driven applications."

Also at Ignite this week

• Microsoft announced it has designed an Azure Boost DPU, apparently its first in-house data-processing unit for accelerating certain workloads in hardware. (There's already some Boost silicon in the Microsoft cloud.)
• It's also revealed what's called the Azure Integrated Hardware Security Module, a security-enforcing chip for its cloud infrastructure.
• Redmond and Atom Computing are collaborating to sell what's claimed to be quantum computers with 1,000+ physical qubits to enterprises sometime next year.
• The Windows maker has listed off ways in which it hopes its OS will avoid another CrowdStrike-like disaster. This includes setting standards for antivirus developers and other software engineers requiring low-level, highly privileged access to the operating system.

Plus...

• Database warhorse SQL Server 2025 goes all-in on AI
• Microsoft unleashes autonomous Copilot AI agents in public preview
• Microsoft goes thin client with $349 Windows 365 Link mini PC

Official announcements here

Fabric Databases would include Azure Cosmos DB and Azure Database for PostgreSQL – already available in Azure – down the line, the vendor said.

Ulagaratchagan said including these services in the Fabric Database would make it "easier for developers to deliver consistent performance even when running spiky workloads, due to features like serverless, auto-scaled compute and storage, and intelligent auto-indexing."

He said the database service would perform both analytical and transactional workloads – usually in separate column and row-oriented systems, respectively – by automatically replicating data to Microsoft OneLake and making it available to the analytical engines within Microsoft Fabric.

While relatively novel, the effort to bring together transactional and analytics workloads is not unique.

Fabric rival Snowflake has launched its Unistore product after previewing the system more than two years ago. Siemens AG is among the customers. Its Hybrid Tables "intelligently identify when a query is transactional or analytical," the vendor said. It stores the data in both row and column formats.

Database SingleStore, which launched in 2019, uses a single representation of the data where the row store is log-structured to tackle the same problem.

Microsoft's Ulagaratchagan said: "We have a complete relational store that supports transaction logs along with columnstore indexes, so you can store data in row and column format. Additionally, data is replicated in near real-time to OneLake as delta parquet."

Melting over Iceberg

Microsoft has also announced developments in its handling of open table formats, which allows users to bring their own query engines to data, without having to move it into another system.

Microsoft's native format is Delta, developed by Databricks and now an open source project managed by the Linux Foundation. In January last year, Microsoft said it would also support rival formats Iceberg and Hudi – both Apache projects – externally.

Last week, Microsoft said customers could now consume Iceberg-formatted data across Microsoft Fabric without data movement or duplication using OneLake shortcuts. It also said Snowflake added the ability to write Iceberg tables directly to OneLake, making it even easier for customers to use both Snowflake and Fabric. "We have nothing new to share at this time regarding Hudi," Ulagaratchagan said.

However, he said Delta would remain the native format, despite AWS, Cloudera, Google, and Snowflake all supporting Iceberg as their preferred format. Last month, former Apple software engineering manager Russell Spitzer – now with Snowflake – said he expected Iceberg to become the dominant format as it gathers vendor support.

"Fabric was optimized from the ground up for the Delta Parquet format," he said. "When we launched Fabric, we committed to making a completely open platform with full interoperability with our partners so our customers can have the flexibility to do what makes sense for their business. That's the reason we standardized Fabric on the open Delta Parquet format and it's the reason we are ensuring customers can access all data sources in Iceberg format, including the Iceberg sources from Snowflake.

"With our latest announcements, customers will get similar performance from either Iceberg or Delta Parquet formats so we've removed any reason to switch." ®