Tweet
I only see that Chris Mason claims 50% lower bound for utilization and in a followup mail explains and writes that it is a "rough number" (in other words, non-rigorous). Not a result formally derived from the algorithms.
-
-
Wow, sorry I didn't notice I had a response until now.
I said *MOSTLY* without issues. You didn't read my entire post. I ran into some issues with discard. I said the issue was due to enabling the discard option in bcache, and that I worked around it with fstrim, which I decided was just the better solution anyway. But to expand on that, I personally suspect the discard issues weren't bcache specific, but rather specific to the discard code in the kernel itself at that time.Originally posted by Licaon View Post
It is great, but it just happens that my other home server is FreeBSD, and BTRFS just isn't an option there. In fact, I backup that system up nightly to my Linux/BTRFS system.Not sure if trolling or something is definitely off here...
Did another person start writing this last paragraph?
Didn't you say it's great? If so why not have backups on BTRFS? Or why not use ZFS since you actually use it where it counts (backups)?
To all that run BTRFS "at home" and "it's great", you also did bought a RedHat license right? If not... if the actual customer is Enterprise, that value little things like reliability and RAIDs and stuff, not sure why anyone would wonder why RedHat dropped it...
I came here to find out "why", as I though the RAID issues where fixed "back then", but nvm :-|
Furthermore, you'd be a fool to put all your eggs into one basket. The chances that I'm going to be hit by an OS/filesystem level bug that wipes out all my data on both my FreeBSD/ZFS and Linux/BTRFS systems at the same time is pretty small. What I really need to do next is work on an additional offsite backup solution.
I'm also currently not using BTRFS RAID. I've used it here and there (with RAID1) and haven't had issues, but in this particular case, performance is more important to me, and therefore I'm running RAID10 on a 3ware 9750 controller.
I don't have a Redhat licence; I run Debian. The point that I'm making here is that at least for me, I haven't had issues with BTRFS. While I realize that the experience of any one individual says nothing about enterprise readiness, we can still acknowledge that I also go much further than the average user, as in I have many terabytes of data, my fileserver is connected to my gigabit switch to the 10GBE uplink port with a 10 GBE nic, I run multiple VM's, and use bcache to get around the otherwise lackluster performance that VM's on top of BTRFS usually results in.
I also run BTRFS on my laptop and the family computer. Those have never had any issues.
Anyway, the word around the campfire is that the primary reason for Redhat's decision is that they employ a fairly large team for XFS, where as they only have 2 BTRFS devs. From that perspective, it makes sense that Redhat decided they're in a much better position to provide enterprise level support for XFS and not so much for BTRFS. This is all just hearsay of course, but this is what I heard at work from people with some connections to Redhat. I'm a Site Reliability Engineer for a large company in Silicon Valley, so my coworkers are better connected than average when it comes to this sort of thing, but it's still just hearsay from a random guy on the internet, so definitely feel free take this with a grain of salt.
Comment
-
We need to clean up some misconceptions in this thread.Originally posted by q2dg View Post
Is ZFS possible to use?
Of course. You can put any module you want in your own kernel and use it. Many distros Including SUSE ship ZFS kernel modules you simply install. Easy peasy. Root on ZFS is a little more involved but it's still very possible to do. You can probably create a new ZFS storage pool on a new disk with the distro your already running in less than 5 minutes.
Is ZFS closed proprietary and anti-Linux?
No, ZFS is open source and it's more freely usable than the Linux kernel itself. The CDDL is a file based copy left licence similar to the MPL that licences many projects such as Firefox and LibreOffice. The licence for it is fine.
If the licence is fine why is it "illegal" to use in Linux?
I don't believe it is illegal to use. This depends how you define derivative works. If ZFS is considered to be a derivative works of Linux then it is illegal to combine with the GPL in code base, meaning in the kernel tree itself. A separate module is fine. However may people, myself included believe it would be silly to define it as a derivative works of Linux as it's clearly a derivative work of Illumos. So in the end, it's for lawyers to decide. As users we can use whatever we want the licence doesn't apply to usage.
But Sun made it to kill Linux right?
No, at least not according to the people that developed it. ZFS came from the larger code release of the OpenSolaris project. The licence was chosen because Sun had to include a compatible licence with binary blobs that they required due to contracts with hardware vendors for things such as scsi controllers in Solaris. GPL just wan't an option for them and it was a surprise to them it wasn't included into Linux. This is just a conspiracy theory saying Sun wanted to kill Linux.. maybe they did who knows but the people that actually did the work didn't believe that and Sun was a fairly good player in the open source world, unlike most companies. They gave us Java, OpenOffice, VirtualBox.. lots of stuff came from Sun and they made money selling hardware, not software.
But doesn't Oracle control ZFS?
No, Oracle has no control over OpenZFS found in Linux, FreeBSD and Illumos. Their version is a fork. Oracle actually controls (or did) Btrfs. Work on OpenZFS can not profit Oracle in any way. Oracle holds copy right over Solaris so they can change the licence but OpenZFS's authors hold copy right themselves so if Oracle takes the changes then they must re-open source Solaris as well.
ZFS isn't well supported on Linux, it's changes are only backported from FreeBSD.
Wrong again. ZFS on Linux (ZoL) is in heavy development and new features such as the new encryption module were developed in ZoL first. ZFS runs huge installations such as a 55-petabyte storage array for the Lawrence Livermore National Laboratory on Linux. This is not a cache disk, this is mission critical scientific data.
End note here if you want to use ZFS on Linux and it fits your needs there is no reason not to.Last edited by k1e0x; 28 August 2017, 04:18 PM.Comment
-
More.
Is ZFS right for everything?
No, it has it's use cases like anything else. It can do a lot of things well but it was intended to be an enterprise file system for big iron systems and for that it does a very good job.
But I need a ton of ECC ram for ZFS?
No you really don't. It depends on the size of your pool and various factors and there are guides out there to help you with this but even if you are below specs all that will hamper is your performance. Dedup is an exception to this as it does take a ton of ram. As far as ECC goes if your in enterprise then you should be using ECC anyway. If your not and your a home user, Non-ECC is ok. Matthew Ahrens one of the core designers of ZFS at Sun has said in an interview "There is nothing special about ZFS that requires/encourages the use of ECC RAM more so than any other filesystem."Comment
-
OpenSUSE says something about "putting butter on a fish" ?!
Comment
-
Your own reference link says that RAID1 and RAID10 are "mostly ok" with the mention "Needs at least two available devices always. Can get stuck in irreversible read-only mode if only one device is present."Originally posted by waxhead View Post
Sorry, but that is not something I would call "rock solid". I have been considering both ZFS and BTRFS and I wanted to go with BTRFS because ZFS claims to need at least 8 GB RAM and BTRFS is built into the kernel and has offline deduplication, but without proper RAID1, RAID10, or RAID5/6 I see it only useful in single disk configurations which is not that appealing.
Maybe Red Hat does not think these multi-disk configuration issues will be fixed any time soon and most likely their customers do not want to run without redundancy.Comment
-
Ok, so let me try to explain the RAID1 part a bit.Originally posted by Alecz View Post
On your average hardware raid or md raid you need to remember that you are only protected against disk failure e.g. not data coruption.
The difference between hw-raid / md raid to btrfs is that btrfs will need two working devices always. If you drop below two devices there is a bug in btrfs that will give you one attempt to replace the broken disk, if you fail to do this the filesystem will go to read-only mode. Since BTRFS is copy on write it can't recover from this since it needs to write to the filesystem (which now consist of one disk) to let it know that it has two working disks.
While this in very inconvenient it is simple to work around. Ensure that you have two working devices always (so run RAID1 with tree or more devices) or be prepared to copy your files to a new disk, add the surviving disk to that and run a rebalance. In any case you will not loose your data and if some file is corrupt you will know what file it is. This is not possible with hw raid / md raid. So if you consider this I would call BTRFS RAID1 rock solid yes, it will give you better protection than hw/md RAID. BTRFS RAID5/6 is a different story. You would be much better off using MD RAID for now.
In the btrfs world RAID1 means TWO COPIES on DIFFERENT DEVICES regardless of how many devices you have. Actually the use of the RAID terminology for BTRFS was never a bright idea in the first place as this lead people to believe that RAID0/1/10/5/6 on BTRFS is the same as for hw-raid / md-raid which it is not. I think that just this is one of the reasons red hat does not care - too many customer complains because they misinterpret what RAID1 actually is.
If you still do not trust BTRFS RAID1 (which have saved me from a few corruptions and a few disk failures as well) then by all means, run MD RAID below and add BTRFS on top. Keep in mind that this will NOT give you any protection against data corruptions (and this goes for any normal filesystem on top of MD RAID) except if you run BTRFS with the DUP data+metadata profile on top which will reduce your available space quite significantly.
http://www.dirtcellar.netComment
-
You should read a little more before writing stupid answers hereOriginally posted by pal666 View PostComment
-
stupid advice from someone who failed to read one directory listingOriginally posted by q2dg View Post
Comment
-
Y'know what is actually stupid? Only reading one directory listing and using that as the sole basis as to whether a filesystem exists. ZFS exists, it just isn't open-source (and therefore not in the mainline kernel). Not only was q2dg's initial question a good one to ask, but his/her advice is ever so relevant.Originally posted by pal666 View Post
Comment
Comment