Windows / Linux Filesystem limits
- Windows: use NTFS for OS / general-purpose / broad compatibility; use ReFS for large data volumes, Hyper-V, and Storage Spaces Direct (S2D) where resiliency and virtualization-oriented performance matter. NTFS is still the default Windows file system, while Microsoft positions ReFS for performance-sensitive and virtualized workloads and explicitly recommends ReFS for S2D volumes.
- Linux: use XFS for large, parallel, throughput-heavy workloads; use ext4 for balanced general-purpose workloads where you want simplicity and broad familiarity. Red Hat recommends XFS as the default local file system unless there is a specific reason not to, while also noting that ext4 can rival XFS on performance in many cases
1) Filesystem limits (high-level, useful version)
Windows
NTFS
- Default file system on modern Windows and supports features like ACLs, BitLocker, quotas, and rich metadata.
- On Windows Server 2019+ / Windows 10 1709+, NTFS volumes can be as large as 8 PB; earlier versions support up to 256 TB. The actual limit depends on cluster size and the maximum number of clusters NTFS supports.
- Microsoft’s filesystem behavior spec says NTFS uses a default cluster size of 4 KB, with a maximum cluster size of 2 MB on Windows 10 v1709+ / Windows Server 2019+ (older versions capped at 64 KB).
ReFS
- ReFS is designed to scale efficiently to very large datasets, provide integrity checks, and perform online repair when paired with Storage Spaces.
- Microsoft support guidance states ReFS supports larger volumes than NTFS, up to 35 PB depending on Windows Server version. Treat that as a practical Microsoft-stated deployment limit, while always checking your exact Windows/cluster/storage design.
- Microsoft’s filesystem spec says ReFS uses 4 KB as the default cluster size on Windows 10+ / Server 2016+, and it also supports 64 KB clusters.
Linux
ext4
- In the Linux kernel documentation, with the common 4 KB block size, ext4 supports a maximum file size of 16 TiB when using extents. Without the
64bitfeature, a 4 KB-block ext4 filesystem is 16 TiB; with the64bitfeature, the theoretical filesystem size becomes far larger. - The Linux kernel docs also note that ext4 typically uses a 4 KB block size, and block size is chosen at
mkfstime. - Red Hat’s docs highlight ext4 as a scalable general-purpose file system and note it supports large files and improved allocation features such as delayed allocation, multi-block allocation, and stripe-aware allocation.
XFS
- Red Hat describes XFS as a highly scalable, high-performance, robust, mature 64-bit journaling file system that supports very large files and file systems.
- In RHEL 10, Red Hat documents supported XFS filesystem size up to 1024 TiB.
- XFS includes extent-based allocation, stripe-aware allocation, delayed allocation, and space pre-allocation, all of which are relevant to high-throughput I/O behavior.
2) What I recommend for best I/O
If you want the clean recommendation matrix
Windows
- Use NTFS if the volume is:
- your OS / boot / general-purpose server volume, or
- you need the broadest compatibility and feature coverage in Windows. NTFS is Microsoft’s default general-purpose choice.
- Use ReFS if the volume is:
- Hyper-V / VHDX-heavy,
- Storage Spaces Direct, or
- a large data volume where resiliency + virtualization-oriented performance features matter. Microsoft calls out ReFS for performance-sensitive and virtualized workloads and recommends it for S2D.
Linux
- Use XFS if the workload is:
- large files,
- high concurrency,
- parallel I/O,
- backup/media/analytics/VM image style throughput workloads.
Red Hat says XFS performs strongly on large systems with enterprise workloads and also for multi-threaded, parallel I/O.
- Use ext4 if the workload is:
- general-purpose,
- mixed,
- or you value simplicity/familiarity and don’t need XFS’s larger-scale focus. Red Hat notes ext4 often rivals XFS on performance.
3) Best-I/O rule that matters more than “filesystem religion”
There is no universally fastest filesystem. The Linux kernel ext4 docs explicitly warn that small changes in workload can completely change which filesystem performs best. In other words: benchmark your actual workload before standardizing.
So the safest best-practice is:
- Start with the platform default that matches the workload
- Windows general-purpose → NTFS
- Windows virtualization / S2D → ReFS
- Linux enterprise local FS → XFS
- Linux simpler general-purpose → ext4
- Benchmark with your real I/O pattern (small random, large sequential, metadata-heavy, VM images, DB files, etc.) because rankings can flip with workload changes.
4) Allocation unit / block size recommendation
Windows
- Stay with the default cluster size unless you have a specific workload reason not to. Microsoft notes that supported NTFS/ReFS volume sizes depend on cluster size, but also warns that changing cluster size can introduce performance issues and may not solve cluster-limit problems.
- On modern Windows, default is 4 KB for NTFS and ReFS, and ReFS also supports 64 KB.
Linux
- For XFS, Red Hat says the default options are optimal for common use.
- For ext4, block size is chosen at filesystem creation and is typically 4 KB.
5) My practical default picks
If I had to give you a one-line recommendation for standard builds:
- Windows server general-purpose / app / file server: NTFS.
- Windows Hyper-V / S2D / large resilient data volumes: ReFS.
- Linux on RHEL / Oracle Linux / enterprise local storage: XFS.
- Linux if you want a simple, familiar, balanced FS: ext4
|
OS |
Filesystem |
Max Filesystem / Volume Size |
Max File Size |
Default Block / Cluster Size |
Best Use Case (I/O Perspective) |
Notes |
|
Windows |
NTFS |
Up to 8 PB (modern Windows); older versions up to 256
TB |
Depends on cluster size (limited by NTFS cluster
architecture) |
4 KB (default); supports up to 2 MB |
General-purpose, OS disks, mixed workloads |
Most compatible Windows FS; full feature set (ACLs,
BitLocker) |
|
Windows |
ReFS |
Up to ~35 PB (practical guidance) |
Not strictly defined in source (very large; designed for
big data sets) |
4 KB default, also supports 64 KB |
Large data volumes, virtualization, high resilience
workloads |
Best for Hyper-V, Storage Spaces Direct; not typically
used for boot |
|
Linux |
ext4 |
~16 TB (32-bit, 4 KB blocks); much higher with
64-bit feature (theoretical up to ZiB scale) |
~16 TB (with extents, 4 KB block) |
4 KB typical |
Balanced workloads, general-purpose Linux |
Stable, widely used; simpler than XFS |
|
Linux |
XFS |
~1024 TB (1 PB) supported (RHEL guidance) |
Up to 8 EB (design capability) |
Typically 4 KB (auto-tuned) |
High-throughput, large files, parallel I/O |
Default in RHEL; excellent for DB, analytics, backups |
| Scenario | Recommended Filesystem |
|---|---|
| Windows OS / system / app server | NTFS |
| Windows large storage / Hyper-V / S2D | ReFS |
| Linux enterprise / heavy I/O / large datasets | XFS |
| Linux general-purpose / mixed workloads | ext4 |
- NTFS = compatibility + general workloads
- ReFS = scale + resilience + virtualization
- ext4 = stable + balanced
- XFS = scale + parallel I/O + large files
The workload pattern (random vs sequential, small vs large files) impacts performance more than filesystem choice.
|
