Other Best filesystem for a shared partition between FreeBSD and Linux

I didn't even know that QFS was still alive. During the late 90s and early 00s, there was a plethora of file systems being developed. HP used Veritas' VxFS, and both together put lots of effort into making it faster (it was very good, in particular when using a database such as Oracle on top of it). SGI was developing XFS, and morphed it into CXFS for clusters (a colleague of mine was the lead architect, and the brother of our neighbor an engineer on it). IBM was morphing the old IBM journaling JFS into JFS2, which also made it into the open source arena. That was a really stable and reliable file system. In addition, IBM had multiple cluster file system projects, GPFS and StorageTank; given the scale of IBM's customers, those distinguished themselves by speed and efficiency (if you want to create a billion files in a million directories, this is what you want to use). Google had published information about GFS (which was radically different, and quite radical), which later begat the Hadoop File System. Sun was quite behind initially (they were still using UFS in Solaris), and they made a massive catch-up effort with QFS and ZFS. QFS vanished around 2008, and ZFS ... we all know how that story ended: happy. Microsoft was, as usual, opaque; but they didn't have to do much, with NTFS and SMB they had extremely good file systems and file sharing protocols, so why waste time on tinkering. In open source, there was the extX camp (ably led by Ted Ts'o, which is probably the best all-around file system and a success story in making sensible compromises), there was Reiser (most known for murdering wives and ab-using ukrainian programmers), and there was the crazy BtfFS group (started at a lunchtime conversation, when Ohad came up with modifyable B-trees). Finally, in academia half the people were madly in love with log-structured file systems (or perhaps they were in love with Margo), the other half were doing planet-wide storage a la Kubi (OceanStore) or Garth (NASD, pronounced nasty). Much fun was had.

And where are we today? We can't even find a good file system that's readable and writeable reliably by both Linux and BSD. EDIT: Other than ZFS, as Eric Borisch pointed out above and below!

Promoting open source? Thank you, but I'd rather promote good engineering products, independent of how the developers are paid.
 
At the risk of being a broken record... ZFS!

Seriously, once you've tried it -- and moved an entire raidz-N pool between operating systems and controller card brands without any issues, or easily recovered that file you accidentally deleted, or looked at the history of changes on /etc/rc.conf (or any other file) without manual version control, or cloned a filesystem instantly to test something out, or used boot environments to perform isolated and revert-able upgrades, or backed up in well under a minute rather than hours -- you'll wonder why you ever used something else.

Recommending anything else (other than the corner cases I mentioned previously) is (IMHO, as everything on a forum is) a disservice, as changing lanes later is always more painful.
 
Personally, I would not trust writing any file system using an implementation that's not done professionally by the first parties....
Would it be acceptable then to pass files back and forth on a raw partition using tar? With tar there is no need for a file system, just a tape-like device -- an unformatted partition will do.
 
For a single group of files, that works great. Even preserves attributes (mtime, ownership) if desired. If one use sensible block sizes, it's even very fast and efficient. The default 10K is not a good match to modern disk sector sizes, but tar works fine with for example 64K or 1M blocks.
 
Would it be acceptable then to pass files back and forth on a raw partition using tar? With tar there is no need for a file system, just a tape-like device -- an unformatted partition will do.
-p can even pass permissions if needed.
 
Back
Top