Storage Spaces Direct

---
name: storage-spaces-direct
description: "Windows Storage Spaces Direct (S2D) hyper-converged storage: Software Storage Bus, SBL cache, storage pools, virtual disks, ReFS, CSV, fault domains, RDMA/SMB Direct, Hyper-V integration, nested resiliency, and Azure Local. Use for \"Storage Spaces Direct\", \"S2D\", \"Azure Stack HCI\", \"Azure Local\", \"ReFS\", \"CSV\", \"Cluster Shared Volume\", \"Software Storage Bus\", \"nested resiliency\", \"mirror-accelerated parity\", \"S2D cache\", \"Get-PhysicalDisk\", \"Get-VirtualDisk\", \"HCI Windows\", \"Hyper-V storage\". Do NOT use for general Windows Failover Clustering not involving S2D storage (quorum, CSV without S2D, cluster networking) — that's the `failover-clustering` skill in the `os` plugin."
license: MIT
---

# Storage Spaces Direct

This skill covers Windows Storage Spaces Direct (S2D), the software-defined storage engine in Windows Server Datacenter and Azure Local (formerly Azure Stack HCI), including:

- Software Storage Bus: software-defined storage fabric over Ethernet replacing physical SANs
- Storage Bus Layer (SBL) cache: server-side NVMe/SSD caching for write absorption and read acceleration
- Storage pools: unified pool from all cluster drives (up to 4 PB, 16 nodes)
- Virtual disks: two-way mirror, three-way mirror, mirror-accelerated parity, nested resiliency
- ReFS: mandatory filesystem with integrity streams, VHDX acceleration, real-time tiering, dedup (WS2025+)
- CSV: Cluster Shared Volumes unified namespace (`C:\ClusterStorage\`)
- Fault domains: site, rack, chassis, node awareness for data placement
- SMB Direct / RDMA: iWARP and RoCE for low-latency inter-node storage I/O
- Hyper-V integration: hyperconverged VMs, live migration, Storage QoS
- Azure Local (Azure Stack HCI): Azure Arc integration, Lifecycle Manager, cloud-based monitoring
- Performance history (WS2019+): 1-year metric retention via `Get-ClusterPerf`
- Windows Server 2025: thin provisioning, NVMe-OF initiator, ReFS dedup/compression, rack-local reads

For cross-platform storage questions, see the `overview` skill.

## How to Approach Tasks

1. **Classify** the request:
   - **Troubleshooting** -- Load `references/diagnostics.md` for PowerShell commands, drive/pool/virtual disk health states, repair workflows, event logs, performance counters
   - **Architecture / design** -- Load `references/architecture.md` for full stack layers, SBL cache, pool, virtual disks, ReFS, CSV, fault domains, RDMA, Hyper-V, Azure Local
   - **Best practices** -- Load `references/best-practices.md` for hardware selection, network design, cache configuration, volume sizing, fault domains, monitoring, maintenance

2. **Identify platform** -- Key differences:
   - Windows Server S2D: hyperconverged + converged (SoFS), manual deployment, on-premises only
   - Azure Local (23H2+): hyperconverged only, Azure Arc mandatory, cloud-managed deployment/updates

3. **Identify version** -- Feature-gated capabilities:
   - WS2016: core S2D, 1 PB max
   - WS2019: nested resiliency, performance history, 4 PB max, 400 TB/server
   - WS2022: SMB east-west encryption, SMB compression
   - WS2025: thin provisioning, NVMe-OF initiator (+90% IOPS), ReFS dedup/compression, rack-local reads
   - Azure Local 22H2: thin provisioning, volume resiliency change, Storage Replica compression
   - Azure Local 23H2: Azure Arc OS, Lifecycle Manager, cloud deployment

## Core Architecture

```
[Networking]      Physical NICs (10 Gbps min; 25 Gbps+ RDMA recommended)
[Clustering]      Windows Failover Cluster
[Storage Bus]     Software Storage Bus (software-defined SAN fabric over Ethernet)
[Cache]           SBL Cache (NVMe/SSD write-back cache for capacity tier)
[Pool]            Single unified storage pool (all drives, all nodes)
[Spaces]          Virtual disks with resiliency (mirror/parity)
[ReFS]            Resilient File System (checksums, VHDX acceleration, tiering)
[CSV]             Cluster Shared Volumes (C:\ClusterStorage\)
[Hyper-V]         VMs on CSV volumes (hyperconverged)
```

### Resiliency Options

| Type | Copies | Efficiency | Min Nodes | Use Case |
|---|---|---|---|---|
| Two-way mirror | 2 | ~50% | 2 | 2-node clusters |
| Three-way mirror | 3 | ~33% | 3 | All-flash performance |
| Mirror-accelerated parity | 1.5-2x | varies | 4+ | General purpose, file servers |
| Nested two-way mirror | 4 | 25% | 2 (exactly) | 2-node max resilience |
| Nested MAP | varies | 35-40% | 2 (exactly) | 2-node balanced |

### SBL Cache Logic

| Media Present | Cache | Capacity |
|---|---|---|
| NVMe + SSD | NVMe | SSD |
| NVMe + HDD | NVMe | HDD |
| SSD + HDD | SSD | HDD |
| All same type | Disabled | All drives |

### Key Limits

| Resource | WS2016 | WS2019+ |
|---|---|---|
| Max pool | 1 PB | 4 PB |
| Max per server | 100 TB | 400 TB |
| Max nodes | 16 | 16 |

## Critical Rules

- Define fault domains BEFORE enabling S2D — data does not retroactively redistribute
- Run `Test-Cluster` before deployment and after hardware changes
- Use storage maintenance mode before node patching
- All SSDs must have power-loss protection (no consumer SSDs)
- RAID controllers must be in HBA pass-through mode
- Number of capacity drives must be whole multiple of cache drives
- All servers: same manufacturer, model, drive configuration

## Reference Files

- `references/architecture.md` -- Full S2D stack, SBL cache, storage pool, virtual disks, fault domains, SMB Direct/RDMA, ReFS, CSV, Hyper-V integration, Azure Local
- `references/best-practices.md` -- Hardware selection, network design, cache configuration, volume sizing, fault domain planning, Windows Admin Center monitoring, performance tuning, maintenance
- `references/diagnostics.md` -- PowerShell diagnostic commands, drive/pool/virtual disk health states, repair workflows, event log analysis, performance counters, common scenarios

## Diagnostic Scripts

Ready-made PowerShell (run elevated on a cluster node) in `scripts/`, numbered by investigation order. All read-only.

- `scripts/01-pool-and-disk-health.ps1` -- Subsystem, pool, virtual disk, physical disk, health faults
- `scripts/02-repair-jobs-status.ps1` -- Repair/rebalance job progress (the maintenance gate)