Return to



Im most likely not the first person to think file systems need a standard. Networking got one.

I loath file systems are all not complete and all competing. Well fat is complete. My meaning was the next true file system.

1 Like


Standard interface or standard implementation?

I think the former is important, but my understanding is that with the exception of fuse stuff, that’s mostly handled by kernel drivers…

The latter, that leads to boxing yourself in and not being able to develop the best option.

The reason nothing is “complete” is because we have reliable systems like EXT4, XFS and UFS. They’re all pretty solid and work for the average Joe.

I think the reason we have so many incomplete filesystems is that we’re in what seems like a transitional stage where we’re moving towards more Software-Defined storage on the enthusiast and professional area.

Think about this: A decade ago, there was no official OS-level way to pool disks together in a way that Windows could see and have insight into. Storage Spaces, for all it’s shortcomings, provided that to us in 2012 R2 and I think windows 8.

I guess my point is that as needs evolve, so does software. We just haven’t reached 100% maturity yet. (arguably, we never will)

That said, we do have some awesome choices.



Well said :slight_smile:

Just the fact that redhat didn’t have BTRFS devs means they are making another new FS now Stratis.

Being redhat it will most likely be paid through to completion. How long to wait sucks.



Stratis isn’t, to my knowledge, a filesystem, but more of an orchestration tool to provide features that filesystems like EXT4 don’t have.

This isn’t a bad thing in itself, it just means that ZFS and BTRFS have the advantage of every component being explicitly designed to work to support all features of the filesystem.

It’s 1.0, as of Oct 2018.

1 Like


Not one mention of xfs and it still needs features to catch up.

How does Stratis compare to ZFS and Btrfs?
In terms of features, Stratis 1.0 does not yet implement some that ZFS and Btrfs have, such as RAID and send/receive support. Not surprising, given their head start in development time.

In terms of its design, Stratis is very different from the two, since they are both in-kernel filesystems. Stratis is a userspace daemon that configures and monitors existing components from Linux’s device-mapper subsystem, as well as the XFS filesystem. This approach makes some things harder to achieve for Stratis, usually involving integration between block-based management and filesystem implementation. For example, the thin-provisioning layer and the XFS filesystem each make no assumption that it is being used along with the other. They are two independent components that Stratis is using together.

But, there are two positives from this difference. First, this lets Stratis development advance more quickly because we does not have to reimplement these components from scratch, just tie them together. These components already were developed, have users, and are maintained and improved on their own. Second, being a userspace daemon makes it easier to perform periodic monitoring tasks, provide an API, as well as potentially integrating with other non-kernel storage-related APIs in the future, such as Ceph, Amazon EBS, or Kubernetes CSI.



It’s only been 2 years. Give it time.

Also, xfs is mentioned in the faq.

It’s 2am. I’ll have to pick this up in the morning.

1 Like


My point was we need a standard like networking. File systems need a standard and not several forks over the next decade.

Stratis is just one of the contenders with from wiki. By the way Stratis is not even yet in the wiki.

Disk file systems are usually block-oriented. Files in a block-oriented file system are sequences of blocks, often featuring fully random-access read, write, and modify operations.

ADFS – Acorn’s Advanced Disc filing system, successor to DFS.
AdvFS – Advanced File System, designed by Digital Equipment Corporation for their Digital UNIX (now Tru64 UNIX) operating system.
APFS – Apple File System is a next-generation file system for Apple products.
AthFS – AtheOS File System, a 64-bit journaled filesystem now used by Syllable. Also called AFS.
BFS – the Boot File System used on System V release 4.0 and UnixWare.
BFS – the Be File System used on BeOS, occasionally misnamed as BeFS. Open source implementation called OpenBFS is used by the Haiku operating system.
Btrfs – is a copy-on-write file system for Linux announced by Oracle in 2007 and published under the GNU General Public License (GPL).
CFS – The Cluster File System from Veritas, a Symantec company. It is the parallel access version of VxFS.
CP/M file system — Native filesystem used in the CP/M (Control Program for Microcomputers) operating system which was first released in 1974.
DOS 3.x – Original floppy operating system and file system developed for the Apple II.
Extent File System (EFS) – an older block filing system under IRIX.
ext – Extended file system, designed for Linux systems.
ext2 – Second extended file system, designed for Linux systems.
ext3 – A journaled form of ext2.
ext4 – A follow up for ext3 and also a journaled filesystem with support for extents.
ext3cow – A versioning file system form of ext3.
FAT – File Allocation Table, initially used on DOS and Microsoft Windows and now widely used for portable USB storage and some other devices; FAT12, FAT16 and FAT32 for 12-, 16- and 32-bit table depths.
VFAT – Optional layer on Microsoft Windows FAT system to allow long (up to 255 character) filenames instead of only the 8.3 filenames allowed in the plain FAT filesystem.
FATX – A modified version of Microsoft Windows FAT system that is used on the original Xbox console.
FFS (Amiga) – Fast File System, used on Amiga systems. This FS has evolved over time. Now counts FFS1, FFS Intl, FFS DCache, FFS2.
FFS – Fast File System, used on* BSD systems
Fossil – Plan 9 from Bell Labs snapshot archival file system.
Files-11 – OpenVMS file system; also used on some PDP-11 systems; supports record-oriented files
Flex machine file system
HAMMER — clustered DragonFly BSD filesystem, production-ready since DragonFly 2.2 (2009)[1][2]
HAMMER2 — recommended as the default root filesystem in DragonFly since 5.2 release in 2018[3][4][5]
HFS – Hierarchical File System in IBM’s z/OS; not to be confused with Apple’s HFS. HFS is still supported but IBM’s stated direction is zFS.
HFS – Hierarchical File System, in use until HFS+ was introduced on Mac OS 8.1. Also known as Mac OS Standard format. Successor to Macintosh File System (MFS) & predecessor to HFS+; not to be confused with IBM’s HFS provided with z/OS
HFS+ – Updated version of Apple’s HFS, Hierarchical File System, supported on Mac OS 8.1 & above, including macOS. Supports file system journaling, enabling recovery of data after a system crash. Also referred to as 'Mac OS Extended format or HFS Plus
HPFS – High Performance File System, used on OS/2
HTFS – High Throughput Filesystem, used on SCO OpenServer
ISO 9660 – Used on CD-ROM and DVD-ROM discs (Rock Ridge and Joliet are extensions to this)
JFS – IBM Journaling file system, provided in Linux, OS/2, and AIX. Supports extents.
LFS – 4.4BSD implementation of a log-structured file system
MFS – Macintosh File System, used on early Classic Mac OS systems. Succeeded by Hierarchical File System (HFS).
Next3 – A form of ext3 with snapshots support.[6]
MFS – TiVo’s Media File System, a proprietary fault tolerant format used on TiVo hard drives for real time recording from live TV.
Minix file system – Used on Minix systems
NILFS – Linux implementation of a log-structured file system
NTFS – (New Technology File System) Used on Microsoft’s Windows NT-based operating systems
NetWare File System – The original NetWare 2.x–5.x file system, used optionally by later versions.
NSS – Novell Storage Services. This is a new 64-bit journaling file system using a balanced tree algorithm. Used in NetWare versions 5.0-up and recently ported to Linux.
OneFS – One File System. This is a fully journaled, distributed file system used by Isilon. OneFS uses FlexProtect and Reed-Solomon encodings to support up to four simultaneous disk failures.
OFS – Old File System, on Amiga. Good for floppies, but fairly useless on hard drives.
OS-9 file system
PFS – and PFS2, PFS3, etc. Technically interesting file system available for the Amiga, performs very well under a lot of circumstances. Very simple and elegant.
ProDOS – Operating system and file system successor to DOS 3.x, for use on Apple’s computers prior to the Macintosh & Lisa computers, the Apple series, including the IIgs
Qnx4fs – File system that is used in QNX version 4 and 6.
ReFS (Resilient File System) – New file system by Microsoft that is built on the foundations of NTFS (but cannot boot, has a default cluster size of 64 KB and does not support compression) and is intended to be used with the Windows Server 2012 operating system.
ReiserFS – File system that uses journaling
Reiser4 – File system that uses journaling, newest version of ReiserFS
Reliance – Datalight’s transactional file system for high reliability applications
Reliance Nitro – Tree-based transactional file system developed for high-performance embedded systems, from Datalight
RFS – Native filesystem for RTEMS[7]
SkyFS – Developed for SkyOS to replace BFS as the operating system’s main file system. It is based on BFS, but contains many new features.
SFS – Smart File System, journaling file system available for the Amiga platforms.
Soup (Apple) – the “file system” for Apple Newton Platform, structured as a shallow database
Tux3 – An experimental versioning file system intended as a replacement for ext3
UDF – Packet-based file system for WORM/RW media such as CD-RW and DVD, now supports hard drives and flash memory as well.
UFS – Unix File System, used on Solaris and older BSD systems
UFS2 – Unix File System, used on newer BSD systems
VxFS Veritas file system, first commercial journaling file system[citation needed]; HP-UX, Solaris, Linux, AIX, UnixWare
XFS – Used on SGI IRIX and Linux systems
zFS – z/OS Distributed File Service zSeries File System; not to be confused with other file systems named zFS or ZFS.
ZFS – a combined file system and logical volume manager designed by Sun Microsystems



they’re block hashes.

how about you go read up on how it works before blindly shitting on the product.



Filesystems have a standard interface with applications, the POSIX VFS. That’s why we can even talk about having different filesystems in the first place.

We have different filesystems because they use different on-disk formats. This is for various reasons - legacy compat (eg ext, ntfs), special applications (eg iso9660, udf).

Take XFS for example. There are multiple implementations of XFS in different operating systems. They talk the same on-disk format. However, there’s a lot of stuff besides understanding a filesystem format that goes into your files landing on the platters.

A filesystem has to talk to disks and to applications somehow. The paths from filesystem land to disk land and to user land go through the operating system’s block storage abstractions and VFS abstractions. These are kernel-specific interfaces that have developed separately over time in different kernels with different needs. The block-storage and POSIX VFS semantics are abstract enough that the same general ideas apply everywhere, but the implementation details differ.

So if you’re wondering why we have different filesystems? Mostly because the filesystems were not developed by the same people and do not have the same goals. The primary goal of ext4 for example is to be backwards compatible with ext2 and ext3. The primary goal of ZFS is to be a modern, highly-reliable filesystem without legacy constraints. The primary goal of HAMMER2 is to support clustering of multiple machines. These are vastly different goals.

To the point that people want a filesystem that works across all the operating systems and isn’t FAT, ZFS is a strong contender. ZFS works on NetBSD, FreeBSD, illumos/Solaris, Linux, macOS, and work is in progress even for Windows. I know there are a few other obscure ones I’m forgetting too (I think OSv for example). Does it scale down to embedded devices very well? Not currently (embedded devices keep getting more and more powerful though). That’s why Apple created APFS instead of using ZFS, in part. Yet another filesystem, with its own unique design constraints.



I appreciate you contributing!



Ty, this is a solid lead :grin:

I was also hoping for a journal to remove the write hole, The idea seems so simple, I wonder why it is not happening.

1 Like


Yeah, the write hole is something that completely slipped my mind.



The parity calculation should be fixed by now, but there is still a write hole which doesn’t appear to have been solved yet in btrfs. ZFS has raid-z, which is designed to eliminate the RAID write hole.

1 Like


I thought the ZIL was also supposed to help with that.



is a zil not a journal?



It is.



yet the write hole persists, Do you have any info on the Zil?
i previously read up on the few years of the kernel on btrfs, but somewhow my mind is blank :<



There was a proposed patch for btrfs that adds an extra disk for a metadata journal as a solution for the write hole, but it doesn’t appear to have been integrated yet.



ZIL is ZFS. ZFS has no write hole.



Oh, that would be sweet. Grab a 32GB optane module and that could help make btrfs really fly.

1 Like