Darryl van der Peijl
Tags:
With Windows Server 2025, Microsoft has just released support for something called an S2D Campus Cluster. Microsoft presents this as a new deployment model for Storage Spaces Direct, aimed at environments where servers are split across multiple racks that are physically separated but still very close together.
The campus cluster changes how Storage Spaces Direct understands fault domains and how it places data, while still remaining a single cluster with a single storage pool.
At its core, Storage Spaces Direct (S2D) is Microsoft’s software-defined storage for failover clustering. It takes local disks from multiple servers, combines them into one big storage pool, and protects your data by replicating it synchronously inside the cluster. From day one, S2D assumed all your nodes were sitting right next to each other. Same room, same rack, very predictable speed. Resiliency is based on that assumption, but it also means S2D was not designed to handle long distances.
Storage Spaces Direct itself was never supported in a “stretched” scenario on Windows Server.
What you could do was cluster-to-cluster replication using Storage Replica. In that scenario, you had two totally separate clusters, each with its own storage pool, and you just copied data from one to the other.
But that is not a true stretched cluster because:
If your primary site went down, the storage replication did not magically move your VMs. You still had to push the button manually or use an external tool to orchestrate the move. It worked, but it was not the seamless “stretched” experience people usually wanted.
The campus cluster takes a different approach. You still have just one S2D storage pool, but your nodes are split across two different racks. The system treats those racks as “fault domains” within the same cluster. We are not using Storage Replica here. S2D itself is doing the work by placing and maintaining copies of your data across those racks. To the cluster, it is all one big system. It is just smarter about where the hardware physically sits.
This is also why Microsoft is careful to call this a “Campus Cluster” rather than a “Stretched Cluster.”
“Campus” is not just marketing fluff here. It describes a specific physical setup where you have separation, but the distance is short enough to keep things fast.
Think of environments like:
Because S2D relies on synchronous replication, the campus cluster does too. Every single write has to be acknowledged by the other rack before it is considered “done.”
That is why the latency requirement is strict. Microsoft recommends a maximum of roughly 1 millisecond round-trip between racks. If you go over that, your write speeds tank, and the cluster can start acting unpredictable. You cannot tune this away. It is just physics and how the storage engine is built.
Campuses can usually handle the latency requirements while still giving you the benefit of surviving a full rack failure.
Campus clusters only support mirror-based resiliency. (Mirror Accelerated) Parity is not an option.
You generally have two choices:
The four-copy option spreads data across racks and nodes. This makes your environment resilient to many failure but you are sacrificing storage space for higher availability.
Especially in Europe, regulations are getting stricter about availability, physical separation, and knowing exactly how recovery works.
Campus clusters are great for this because:
For many regulated industries, this is a much cleaner fit than trying to stretch clusters across cities or relying on slower, asynchronous copying.
The S2D Campus Cluster fills a gap we have had for a long time. It sits right between the classic “everything in one room” setup and the complex “multi-site disaster recovery” designs. As long as your network is fast enough, it is a clean, predictable way to boost availability without making your life overly complicated.
If you are considering an S2D Campus Cluster or want to understand whether it fits your environment, feel free to reach out to us. We work with Storage Spaces Direct and Windows Server clustering on a daily basis and are happy to help you assess the design, the trade-offs, and whether this model actually makes sense for your use case.
Gerelateerde blogs
