Wow, KubeCon + CloudNativeCon Europe 2025 in London was a implausible time! Our Azure Storage workforce was thrilled to trade insights and success tales at this vibrant group occasion. When you missed it, don’t fear! We’re right here to share what we confirmed off on the con: how we’re enhancing efficiency, cost-efficiency, and AI
Wow, KubeCon + CloudNativeCon Europe 2025 in London was a implausible time! Our Azure Storage workforce was thrilled to trade insights and success tales at this vibrant group occasion. When you missed it, don’t fear! We’re right here to share what we confirmed off on the con: how we’re enhancing efficiency, cost-efficiency, and AI capabilities for your workloads on Azure.
Optimize your open-source databases with Azure Disks
Open-source databases equivalent to PostgreSQL, MariaDB, and MySQL are among the many mostly deployed stateful workloads on Kubernetes. For eventualities that demand extraordinarily low latency and excessive enter/output operations per second (IOPS)—equivalent to working these databases for transactional workloads—Azure Container Storage allows you to faucet into native ephemeral non-volative reminiscence categorical (NVMe) drives inside your node pool. This offers sub-millisecond latency and as much as half 1,000,000 IOPS, making it a robust match for performance-critical use instances. In our upcoming v1.3.0 replace, we have now made vital optimizations particularly for databases.
In comparison with the earlier v1.2.0 model, you’ll be able to count on as much as a 5 occasions enhance in transactions per second (TPS) for PostgreSQL and MySQL deployments. When you’re in search of one of the best steadiness of sturdiness, efficiency, and value for storage, Premium SSD v2 disks stays our advisable default for database workloads. Premium SSD v2 gives a versatile pricing mannequin that costs per gigabyte and consists of beneficiant baseline IOPS and throughput free of charge out of the field. You may dynamically scale IOPS and throughput as wanted, permitting you to fine-tune efficiency whereas optimizing for price-efficiency.
At KubeCon, we demonstrated how builders may readily make the most of native NVMe and Premium SSD v2 Disks to construct their extremely out there and performant PostgreSQL deployments. If you wish to comply with alongside your self, try the newly republished PostgreSQL on AKS documentation beneath!
Speed up your AI workflows with Azure Blob Storage
Constructing an AI workflow calls for scalable storage to host large quantities of information—whether or not it’s uncooked sensor logs, high-resolution pictures, or multi-terabyte mannequin checkpoints. Azure Blob Storage and BlobFuse2 with the Container Storage Interface (CSI) driver present a seamless option to retailer and retrieve this information at scale. With BlobFuse2, you entry blob storage simply as a persistent quantity, treating it like an area file system. With the newest model of Blobfuse2 – 2.4.1, you’ll be able to:
- Pace up mannequin coaching and inference: BlobFuse2’s enhanced streaming assist reduces latency for preliminary and repeated reads. Utilizing BlobFuse2 to load giant datasets or fine-tuned mannequin weights immediately from blob storage into native NVMe drives on GPU SKUs successfully optimizes the effectivity of your AI workflow.
- Simplify information preprocessing: AI workflows typically require frequent transformations—equivalent to normalizing pictures or tokenizing textual content. Through the use of BlobFuse2’s file-based entry, information scientists can preprocess and retailer outcomes immediately in blob storage, preserving pipelines environment friendly.
- Guarantee information integrity at scale: When dealing with petabytes of streaming information, integrity checks matter. BlobFuse2 now consists of improved CRC64 validation for information saved on native disk, making certain dependable reads and writes, even when working with distributed AI clusters.
- Parallel entry of large datasets: Applied parallel downloads and uploads to considerably lower the time required for accessing giant datasets saved in blobs. This enhancement permits for quicker information processing and elevated effectivity, making certain optimum utilization of GPU assets and bettering coaching effectivity.
Scale your stateful workloads with Azure Recordsdata
Steady Integration and Steady Supply/Deployment (CI/CD) pipelines, one of many standard stateful workloads, want shared persistent volumes to host artifacts within the repository, the place Azure Premium Recordsdata is the storage of selection on Azure. These artifacts are saved in small information that incur heavy metadata operations on a file share. To hurry up CI/CD workflows, the Azure Recordsdata workforce just lately introduced the overall availability of metadata caching for premium SMB file shares. This new functionality reduces metadata latency by as much as 50 %, benefiting metadata-intensive workloads that sometimes host many small information in a single share. At KubeCon, we showcased how metadata caching can speed up repetitive construct processes on GitHub, confer with the repo and check out it out your self.
For much less performance-demanding stateful workloads, Customary Recordsdata with the brand new Provisioned v2 billing mannequin gives higher price predictability and management as shared persistent volumes. The Provisioned v2 mannequin shifts from usage-based to provisioned billing, permitting you to specify required storage, IOPS, and throughput with larger scale. Now you can increase your file share from 32 GiB to 256 TiB, 50,000 IOPS, and 5 GiB/sec throughput as your functions demand.
KubeCon + CloudNativeCon was a beautiful alternative to immediately work together with builders and be taught from our prospects. As all the time, due to our prospects and companions for contributing to the occasion’s worth and significance, and we sit up for seeing you once more in November for KubeCon North America!