Is NTFS on the way out? [LWN.net]

Is NTFS on the way out?

Posted Jan 24, 2026 14:30 UTC (Sat) by wtarreau (subscriber, #51152)
In reply to: Is NTFS on the way out? by djwong
Parent article: Filesystem medley: EROFS, NTFS, and XFS

> Filesystems are hard, XFS is just about finished, and it's been what, 33 years? 😜

Has it become better with small files ? I was using it a lot in the past, but I had to abandon it on my laptop for ext4 15 years ago when every git operation on my linux directory would make my disk seek for 30 seconds before completing the operation. However for large files, streaming writes etc it was awesome. I initially started to use it on network probes running tcpdump that would continuously capture traffic on multiple interfaces and store it to disk. Writing and rotating 10GB files was as instant as doing it on a readme with XFS, and I'm speaking about pentium4 with sata1 disks. Also it never failed even once on me, despite not having invoked "halt" nor "shutdown" on a linux machine for the last 20 years, since boot-time journal replay is always faster than a "clean" system shutdown.

On the other hand, ext4 is also good for my use cases nowadays, I'm just not sure if XFS would bring me anything more, or could still annoy me on numerous small files like a kernel source tree.

to post comments

Is NTFS on the way out?

Posted Jan 24, 2026 21:28 UTC (Sat) by dgc (subscriber, #6611) [Link] (1 responses)

> Has it become better with small files ? I was using it a lot in the past,
> but I had to abandon it on my laptop for ext4 15 years ago when every
> git operation on my linux directory would make my disk seek for 30
> seconds before completing the operation.

Yes, most of major small file/metadata performance issues (in comparison to ext4) were fixed with the introduction of delayed logging in 2.6.36 (i.e. about 15 years ago). That improved transaction rates from being disk bound at a few hundred a second to being CPU bound at tens of thousands of transactions/second.

These days XFS can run a couple of million asynchronous transactions per second through the journal, so highly concurrent workloads are able to scan, create, rename, unlink, etc at rates of over half a million files/second, even on a typical cheap consumer SSDs....

Is NTFS on the way out?

Posted Jan 26, 2026 5:19 UTC (Mon) by wtarreau (subscriber, #51152) [Link]

Good to know, thank you! I'll have to try it again then!