Slashdot Mirror
Ubuntu 16.04 LTS To Have Official Support For ZFS File System (dustinkirkland.com)
LichtSpektren writes: Ubuntu developer Dustin Kirkland has posted on his blog that Canonical plans to officially support the ZFS file system for the next Ubuntu LTS release, 16.04 "Xenial Xerus." The file system, which originates in Solaris UNIX, is renowned for its feature set (Kirkland touts "snapshots, copy-on-write cloning, continuous integrity checking against data corruption, automatic repair, efficient data compression") and its stability. "You'll find zfs.ko automatically built and installed on your Ubuntu systems. No more DKMS-built modules!" N.B. ext4 will still be the default file system due to the unresolved licensing conflict between Linux's GPLv2 and ZFS's CDDL.
I'll stick with BTRFS thanks. It gives me all those features, is GPL and has been trouble free for me on many TB of disks for several years.
I think you don't know what ZFS really is. It's a very different deal than etx4, ufs etc... It is the file system that made HW raid controllers obsolete.
It also made just about any computer with less than 8 GB of RAM obsolete. It's also not very friendly with applications that need large chunks of RAM, like a database or large Java VM application - the ARC cache causes a lot of fragmentation and is often slow to release it when other applications need more.
"Somebody has to do something. It's just incredibly pathetic it has to be us."
--- Jerry Garcia
It's not quite abandonware, but the central impetus for it's creation and advancement is gone.
I wasn't planning to comment on this thread, but this is too big a lie to let stand -- unless by "not quite abaondonware" you mean "has absolutely nothing in common with abandonware besides being a type of software". Oracle was never the sole developer, and now that Oracle has lost interest, the developers just moved to other companies and kept doing the same thing. Its raison d'etre remains to provide an advanced filesystem that's easily integrated with linux, which for better or worse means being licensed under the GPL or something compatible.
As for encryption, yeah that would be nice to have, but it's not like zfs has all the features btrfs has. I'll take btrfs's online balancing (ability to add and remove drives at will) over built in encryption, but I realize that's a personal choice.
Finally, let's actually quote the FAQ correct only stability:
Short answer: Maybe.
Long answer: Nobody is going to magically stick a label on the btrfs code and say "yes, this is now stable and bug-free". Different people have different concepts of stability: a home user who wants to keep their ripped CDs on it will have a different requirement for stability than a large financial institution running their trading system on it. If you are concerned about stability in commercial production use, you should test btrfs on a testbed system under production workloads to see if it will do what you want of it. In any case, you should join the mailing list (and hang out in IRC) and read through problem reports and follow them to their conclusion to give yourself a good idea of the types of issues that come up, and the degree to which they can be dealt with. Whatever you do, we recommend keeping good, tested, off-system (and off-site) backups.
Pragmatic answer: (2012-12-19) Many of the developers and testers run btrfs as their primary filesystem for day-to-day usage, or with various forms of real data. With reliable hardware and up-to-date kernels, we see very few unrecoverable problems showing up. As always, keep backups, test them, and be prepared to use them.
For all practical purposes, btrfs is stable. Everything they say in the long answer basically applies to linux in general (unless you have a support contract with Red Hat or the likes).
On 64-bit hosts, the ARC cache is a non-issue. Java needs contiguous *virtual* memory space. Physical memory fragmentation isn't a problem w/ the MMU translating contiguous 64-bit address space to possibly non-contiguous physical pages. On 32-bit hosts, that gets dicey. On 64-bit, you've got plenty of room even w/ ARC.
That said, I'd love to see ARC & the native Linux disk cache functionality either merge or at least have ARC behave more like the normal caching mechanism (IE free up RAM more eagerly), but it's not actually caused me significant problems on 64-bit.
I used MythTV for years as a DVR and I tried a lot of different file systems.
The 2 that always worked the best were JFS and XFS for the sole reason that large file deletes took almost no time at all. Compared to several seconds or even minutes with other file systems.
My eyes reflect the stars and a smile lights up my face.
Also don't enable dedup if you have media with a nontrivial seek time. It's tolerable on flash (but you do lots of extra I/O on write) but the deduplication table (DDT) tends to develop a random layout with respect to device LBAs, and the DDT needs to be consulted on each write to a dataset/zvol with dedup enabled, and it also needs to be scanned *first* during scrubs and resilvers. DDTs with millions of entries can require hundreds of thousands of random I/Os, which means hundreds of thousands of seeks. At 10-30 ms/seek, you will have a bad time if you ever need to replace a disk in your pool, or scrub regularly.
Additionally, you have to update the DDT when you remove duplicates. When you remove a file from a deduplicated dataset: DDT consult. If DDT is not in memory, that means random I/O. When you remove a snapshot filled with data entered into the DDT: DDT consults for each record (128k, typically; sometimes smaller). Worse, asynchronous destruction of snapshots and datasets across a reboot boundary must be finished before the pool becomes available during the import process. And at the reboot, your ARC is empty and your L2ARC is unavailable before import, so each removed record is much more likely to result in a disk seek.
On SSDs, not so bad. On rotating media, deduplication can result in operational disasters -- I have personally seen pools take a WEEK to import after an unexpected restart during a "zfs destroy -R foo/bar" where bar has dedup=on.
I personally have read a lot of lost data horror stories with btrfs. Usually it isn't a small chunk of data, but a large partition that winds up with major problems. Because of this, I wouldn't want to use btrfs for anything production.
ZFS, on the other hand, has been tested in a ton of environments, and is what the big boys use when using Oracle/Solaris/SPARC.
Now, if RHEL and downstreams can feature ZFS, life will be nice.