Disaggregated Storage on Azure Local
Hyperconverged infrastructure is brilliant for a huge number of use cases. Combining compute, storage, and networking into a single, scalable building block simplifies procurement, deployment, and management. Storage Spaces Direct in Azure Local delivers outstanding performance, and as I’ve noted previously, you can push north of a million 4K IOPS on a modest two-node cluster. For most workloads, HCI is the right answer.
But not for all workloads. And this is where disaggregated storage comes into the picture, specifically Dell PowerFlex with the Dell AX-4520c.
When HCI Isn’t Enough
There are a few scenarios where the inherent constraints of hyperconverged infrastructure start to bite.
Disproportionate storage requirements. If you need a large amount of storage relative to compute, HCI forces you to buy compute nodes you don’t need just to get the storage capacity. This is wasteful and expensive. A database workload that needs 200TB of storage but only a handful of CPU cores is the classic example.
Scale boundaries. Azure Local supports clusters up to 16 nodes. While 16 nodes of modern server hardware provide a tremendous amount of compute and storage, there are scenarios where you need more storage than 16 nodes can provide without exceeding the cluster size limit.
Independent scaling. In an HCI model, adding storage means adding compute and vice versa. If your storage growth rate is different from your compute growth rate, you end up with an imbalanced cluster over time. Disaggregated storage lets you scale each dimension independently.
Performance isolation. In an HCI cluster, storage I/O competes with compute workloads for the same CPU, memory, and network resources. For workloads with demanding storage performance requirements running alongside compute-heavy workloads, this can create contention. Disaggregated storage offloads storage processing to dedicated infrastructure.
Enter PowerFlex
Dell PowerFlex is a software-defined storage platform that can run on dedicated storage nodes, providing block storage to compute clusters over the network. It’s not new, it’s been used in VMware environments, bare metal deployments, and public cloud scenarios for years. What’s newer is its validated integration with Azure Local through the Dell AX-4520c.
The AX-4520c is specifically designed for disaggregated storage scenarios. It connects to a PowerFlex cluster to provide external block storage to the Azure Local instance, while the Azure Local compute nodes focus on running workloads. The result is a deployment where your compute scales with your compute needs, and your storage scales with your storage needs, independently.
From an architecture perspective, the Azure Local cluster still uses Storage Spaces Direct for its core infrastructure storage (the OS, the Arc Resource Bridge, and other system components). The PowerFlex storage is presented to the cluster as additional storage volumes for workload data. This means you get the best of both worlds: the simplicity of S2D for system infrastructure, and the scalability of PowerFlex for workload storage.
Performance
I mentioned earlier that S2D performance on Azure Local is impressive. PowerFlex performance is in a different league entirely. With dedicated storage nodes running NVMe drives and a high bandwidth storage network, PowerFlex can deliver the kind of sustained IOPS and throughput that the most demanding database and analytics workloads require.
Dell has published performance benchmarks for the AX-4520c with PowerFlex that demonstrate significant gains over S2D for large scale, I/O intensive workloads. The specifics will depend on your configuration, but the general principle holds: dedicating hardware resources to storage allows you to push storage performance further than sharing those resources with compute workloads.
For Dell AX customers, the performance validation is part of the Premier Solution testing programme. PowerFlex configurations are tested end to end with Azure Local releases, ensuring that updates don’t introduce performance regressions.
When to Choose What
The decision between HCI-only and HCI plus disaggregated storage isn’t binary, and it shouldn’t be made based on which sounds cooler. Here’s how I’d think about it.
Start with HCI. For most Azure Local deployments, Storage Spaces Direct provides excellent performance, simplicity, and cost effectiveness. If your workloads fit comfortably within the storage capacity and performance envelope of your compute nodes, there’s no reason to add complexity.
Consider disaggregated storage when you hit a wall. If you’re running out of storage capacity before you’re running out of compute, if you need storage performance that exceeds what S2D can deliver, or if you have workloads that require performance isolation between compute and storage, that’s when PowerFlex starts to make sense.
Think about the future. If you’re deploying today for a workload that’s modest in scale, but you know that in two years the storage requirement is going to grow substantially, it may be worth designing for disaggregated storage from the outset rather than having to re-architect later.
The Multicloud Storage Story
Stepping back, this is part of a broader Dell storage story across the multicloud landscape. PowerFlex is available on-premises with Azure Local, but it’s also available in Azure and other cloud environments. If you’re building a multicloud storage strategy, having a consistent storage platform that works across on-premises and public cloud reduces the complexity of data mobility, replication, and disaster recovery.
For customers who are already invested in PowerFlex in other parts of their infrastructure, extending it to Azure Local through the AX-4520c is a natural next step. For those who are new to PowerFlex, the Azure Local integration provides an on-ramp into a broader storage ecosystem.
Either way, having the option to disaggregate compute and storage on Azure Local opens doors that were previously closed, and for the right workloads, it’s a meaningful capability.