Accelerating AI and databases with Azure Container Storage, now 7 occasions sooner and open supply


The following main launch of Azure Container Storage delivers as much as 7 occasions greater IOPS and 4 occasions much less latency in comparison with earlier variations.

Extra corporations than ever earlier than are selecting to run stateful workloads—comparable to relational databases, AI inferencing, and messaging queues—on Kubernetes. For builders constructing on Kubernetes, storage efficiency has by no means been extra essential.

At present, we’re saying the subsequent main launch of Azure Container Storage – v2.0.0. In comparison with prior variations, it delivers as much as 7 occasions greater IOPS, 4 occasions much less latency, and improved useful resource effectivity. With built-in assist for native NVMe drives, Azure Container Storage now delivers our quickest, strongest Kubernetes storage platform on Azure. It’s now additionally fully free to make use of, and out there as an open-source model for set up on non-AKS clusters. Whether or not you’re working stateful functions in manufacturing, scaling AI workloads, or streamlining dev/take a look at environments, this main launch’s efficiency will give your workloads a substantial increase.

What’s Azure Container Storage?

Earlier than we dive into the newest enhancements, let’s take a second to revisit what Azure Container Storage is and the way builders run stateful workloads on Kubernetes with velocity, simplicity, and reliability.

Azure Container Storage is a cloud-native quantity administration and orchestration service particularly designed for Kubernetes. It integrates seamlessly with AKS (Azure Kubernetes Service) to allow provisioning of persistent volumes for production-scale, stateful workloads.

Azure Container Storage’s imaginative and prescient is to function the unified block storage orchestrator for Kubernetes workloads on Azure, offering a constant expertise throughout a number of storage backends for simplified quantity orchestration by way of Kubernetes APIs. This v2.0.0 launch focuses particularly on breakthrough efficiency with native NVMe storage, bringing enterprise-grade efficiency with cloud-native simplicity. Later this yr, we’ll be debuting assist for Azure Container Storage to combine with Elastic SAN.

Azure Container Storage delivers optimized efficiency and effectivity with low-latency storage for prime throughput stateful functions, together with built-in orchestration and automation that enables Kubernetes to handle storage swimming pools, persistent quantity lifecycles, snapshots, and scaling—all with out switching contexts or managing particular person CSI (container storage interface) drivers.

What’s new?

There’s fairly a bit to unpack right here, so let’s take a deeper dive into among the key advantages that Azure Container Storage v2.0.0 delivers:

Pricing modifications

As earlier than, you’ll proceed to pay for the underlying storage backend you utilize. However Azure Container Storage variations 2.0.0 and past will now not cost a per-GB month-to-month price for storage swimming pools bigger than 5 TiB for each our first occasion managed and open-source model, making the service now fully free to make use of. Provision as a lot storage as you want with out worrying about extra administration charges. This implies you get enterprise-grade storage orchestration and breakthrough efficiency with none extra service prices—simply pure worth to your Kubernetes workloads.

Enhanced efficiency with decreased useful resource consumption

This launch of Azure Container Storage is optimized particularly for native NVMe drives supplied with quite a lot of VM households. This focus unlocks the quickest potential I/O efficiency to your most demanding workloads whereas lowering infrastructure prices.

Maybe most fun, this newest model of Azure Container Storage on native NVMe is now sooner than ever earlier than. We’ve rebuilt our structure from the bottom up—from the kernel stage to the management aircraft—to push the boundaries of our storage orchestrator. This dramatic velocity enchancment comes with an equally spectacular discount in cluster useful resource consumption. Beforehand, Azure Container Storage on native NVMe had three efficiency modes that would devour 12.5%, 25%, or 50% of your node pool’s CPU cores. Azure Container Storage v2.0.0 now not has efficiency tiers. As an alternative, it delivers superior efficiency whereas utilizing fewer sources than even our earlier lowest-impact setting. This interprets on to value financial savings—you get higher efficiency whereas liberating up CPU capability to your functions to carry out even sooner.

Let’s take a look at the benchmarks. On fio (Versatile I/O Tester), the open-source trade customary for storage testing, Azure Container Storage on NVMe delivers roughly 7 occasions greater IOPS and 4 occasions much less latency in comparison with the earlier model.

However how does this translate to actual workloads? We examined our personal PostgreSQL for AKS deployment information and located that PostgreSQL’s transactions per second improved by 60% whereas chopping latency by over 30%. For database-driven functions, this implies sooner question responses, greater throughput, and higher consumer experiences.

Transactions per sec, higher is better.
Latency comparison, lower is better.

All in all, Azure Container Storage delivers a major efficiency increase for I/O-demanding workloads out of the field with out extra configuration wanted, providing builders a easy but highly effective device of their cloud-native arsenal.

Accelerated AI mannequin loading and KAITO Integration

For AI and machine studying workloads, mannequin loading time generally is a vital bottleneck. Azure VMs outfitted with GPUs have native NVMe drives out there. With the newest NVMe enhancements within the new v2.0.0 model, Azure Container Storage takes benefit of this {hardware} by dramatically accelerating mannequin file loading for AI inferencing workloads. With our current integration with KAITO, the primary Kubernetes-native controller for automating AI mannequin deployment, now you can deploy and scale AI fashions sooner than ever, lowering time-to-inference and enhancing total AI software responsiveness.

Diagram of KAITO distributed inference with A C Stor.
Above: Azure Container Storage offering quick NVMe-backed storage for mannequin information

We loaded Llama-3.1-8B-Instruct LLM and located a 5 occasions enchancment in mannequin file loading velocity with Azure Container Storage v2.0.0, in comparison with utilizing an ephemeral OS disk.

A graph of data being used

Extra versatile scaling choices

Azure Container Storage beforehand required a minimal of three nodes when utilizing ephemeral drives. It now works with clusters of any dimension, together with single-node deployments. This flexibility is especially worthwhile for functions with strong built-in replication or backup capabilities, growth environments, and edge deployments the place you want high-performance storage with out the overhead of bigger clusters. The elimination of minimal node necessities additionally reduces prices for smaller deployments whereas sustaining the identical high-performance capabilities.

Open supply and group assist

We acknowledge how essential the open-source group is to the well being and spirit of the Kubernetes ecosystem. Azure Container Storage model 2.0.0 is now constructed on our newly created open-source repositories, making it accessible to the broader Kubernetes group.

Whether or not you want the Azure-managed model for seamless AKS integration or choose the group open-source model for self-hosted Kubernetes clusters, you get the identical nice product and options. The open-source strategy additionally means simpler set up, higher transparency, and the flexibility to contribute to the undertaking’s evolution.

Discover our open-source repository (local-csi-driver), and be taught extra about our associated block storage merchandise:

In abstract

This main replace to Azure Container Storage delivers a sooner and leaner high-performance Kubernetes storage platform. Right here’s what you get:

  • Included out of the field: This launch focuses on ephemeral drives (native NVMe and momentary SSD) supplied with choose VM households, together with storage-optimized L-series, GPU-enabled ND-series, and general-purpose Da-series.
  • Enhanced workload assist: Optimized for demanding functions like PostgreSQL databases and KAITO-managed AI mannequin serving.
  • Superior efficiency: 7 occasions enchancment in learn/write IOPS and 4 occasions discount in latency, with 60% higher PostgreSQL transaction throughput.
  • Open supply: Constructed on open-source foundations with group repositories for simpler set up on any Kubernetes cluster.
  • Versatile scaling: Deploy on clusters with as few as one node—no minimal cluster dimension necessities.
  • Zero service charges: Utterly free to make use of for all storage pool sizes—you solely pay for underlying storage.

Getting began

Able to expertise the efficiency increase? Listed below are your subsequent steps:

No matter your workload, Azure Container Storage gives the efficiency and ease you count on from trendy cloud-native storage. We’re excited to see what you construct—and we’d love to listen to your suggestions. Pleased hacking!