ZFS on Linux the Billion dollar file system

In this episode of the CyberGizmo we explore ZFS on Linux (ZoL). I wanted to cover this now because Ubuntu 19.10 will have an experimental mode to use ZFS as the root file system. I am by no means a ZFS expert I have used it in a massively large file store in production and switched over to using it at home about 2 years ago. I use BSD for my ZFS, so I am learning the linux side of things with you.

I cover the commands you need to get started, how to add on additional storage, troubleshooting failures, replacing drives, improve performance and export your data to any system running ZFS

Follow me:
Twitter @djware55

Music Used in this video
Licensed under Creative Commons: By Attribution 3.0 License

[dragontav]

Good information. Please consider not using an animated background. The text flying by was distracting since I was trying to read along in your shell.

[AlexChambersXYZ]

Quality video. I like this format and teaching style. Thank you for this video! Would nice to see a Samba / NFS crash course from you

[ricebowl___]

Thanks for the great ZoL overview. I have a bit more confidence managing ZFS pools and datasets now.

[drkskwlkr]

I enjoyed immensely the format of this video, and subscribed to the channel immediately afterwards. I have a feeling this will become one of my favorite YouTube channels!

[DanielPeraalta]

Much obliged for this amazing class, so useful.. I got rid of many questions. But this transfer speed though (12:31) such an eye candy!

[BassBastiforever]

Thanks a lot for the easy explained basics of zfs! I'm going to my build a NAS Server for my flat and I'm planning to use TrueNas.

[demerit]

Another great video. Thank you for the upload

[DanCalloway]

Nice. Great video. Lots of info. Just for your edification, the command to display the QUOTA assigned to djware @ 25:24 in the video should have been: zfs get quota zfspool/djware. Replace ALL with QUOTA and it will list out only the quota for djware and not all status.

[prashanthb]

Clear demo & explanation. Thanks a lot sir.

[faziz]

Thank you for to the point intro to ZFS. Please do a video on Samba sharing with ZFS. Thanks.

[lsatenstein]

Hi DJ.
First of all, today, inauguration day, the USA has a new captain to steer the boat. The choice of captain and co-captain do know how to avoid the rocks, and where it is safe to dock.

I am following up on a earlier email message from you that I took to heart and did something. I installed a version of Ubuntu 2010 with zfs .

After watching your zfs on Linux Billion dollar question, I allocated a 120gig SSD to Ubuntu 2010 ZOL. My experience has been different from day to day.

On the first reboot, I had to install software that I use daily (clang, meld, vim, and some others). Each use of apt install/upgrade automatically generated a new sub-volume.,
Thus far I have about 30 sub-volumes and have not discovered a zfs method to eliminate the older ones.

The first few days, as I did my testing, the system performance was slow but improving. zfs has a very large cache and manages it as a repository for short term and long term management. I takes time to fill that cache to where a probe into it is rapid.

A program that I use over several days tends to reside almost permanently within the cache, while a reload or two of the same program within some hours of each other offers benefits for the second and later program reloads. Without reading the source code, I will assume that the cache also serves to hold some data in it's store. (config files, etc). During my evaluation, Ubuntu provided some linux software and some zfs updates. These updates were bug fixes, and not performance fixes. After a few days, system response time was slower than if I used btrfs, but good enough to consider remaining with zfs, , given zfs's crash and other recovery abilities. In terminal mode, a file save appears to occur some 1/4 second after the actual file was closed.

After a week of UBUNTU 2010 ZFS use, I have to say that there were only a few issues with my Ubuntu 2010 and zfs installation. These are:
a) The terminal handler for the keyboard goes through the file system, which results in the keyboard input, appearing on the screen only after some delay of perhaps 1/4 second. I can type the letter a, and wait to watch it appear some time after I entered the character. UBUNTU 2010 with zfs is based on the current gnome version. I noted nothing special about the Gnome installation.
b) another quirk about the UBUNTU 2010 installation was the blocked - non-ability to log into the system directly, by foregoing the GUI interface. This access method is blocked or does not exist, or is broken. When trying direct terminal access the gui software shuts down---passwords are refused. However, if I log into the system, and then enter terminal mode, this route works fine.
c) Updating grub fails, to replace an original grub.cfg. The latter, created at time of Ubuntu installation remains in effect. Updates to grub.cfg do not seem to take effect. My proof: "I had removed a Linux that was installed alongside of Ubuntu, and I was unable to update grub to show that the other distro was no longer installed". It's entry remained within the grub menu.
d) I was able to install my brother printer and print a test page. So, printing is working.
e) I have not been able to remove older zfs snapshots, and I feel that the ones that were created at time or installation should be purge-able .
f) I did like the fact that my user number was 1000, with default group being 1000, consistent with Fedora, and Centos. Sudo setup works as expected.

So, here is my opinion about zfs on Ubuntu2010 Linux. Ubuntu2010 with zfs is not yet ready for prime time. It is ok to use if all you do is desktop work or games. There is really no emergency access by direct terminal mode if a need arose. I believe that btrfs is better suited to the desktop than is zfs.

What I would like to see in zfs for the future.
a) To be able to setup by partition, rather than by disk. As I do not store gigabytes or terrabytes of files on my desktop system two 50 gig partitions would suffice. The most I have ever stored on my active distribution is 10 gits of data. I would have liked to partition my 120gig SSD tUbuntu2010 installation, into two parts of 55 gigs, and use each part as a raid pair. I did not see anywhere in the literature where I could fake a partition into emulating a physical disk. (aside, yes, separate disks for raid).

In closing. I appreciate your youtube presenations and commentary and your discussion as a prelude to your physical installation demonstration. I look forward daily to your you-tube videos. You fill a void that is not covered by other presenters of Linux/bsd related software.

[eznix]

Thanks for the intro into ZFS on Linux. ZFS looks like overkill for my home use, but I can see why there is such interest.

[davidinvenio]

Great video, very useful and as always very professional and easy to listen to and watch (so happy to read you won't do the animated backgrounds anymore although I've already seen newer videos and would have noticed it - and quick question about backgrounds, on your main workstation do you use transparency for your terminal windows etc.? or was it just for the video? I could never use it myself and wondered what the benefit was!)
Thanks again for another great video and I hope you don't take my criticisms too seriously! Your videos are far far better than the vast majority here! Far far far far far :-)

[richardbennett]

It's for when deduplication is set to On. When deduplication for a dataset is on, then it is said for a rough rule of thumb one needs another GB RAM for each TB reduplicated. And, it's worse than that, because ZFS only allows 25% of RAM to be used for metadata, and the deduplication table that's held in RAM is part of / considered metadata, so multiply by 4 to end up with 4 GB RAM for V1 TB DEDUPLICATED data.

[edwardg]

Thanks a lot. Good pace and very great graphics too :)

[johnware]

Really interesting! I was looking for something like this when my file server was hosted with Linux Mint. It's not up yet, but it will be using FreeNAS since I can integrate it with my Active Directory server.

[themistoclesnelson]

I enjoyed the video and feel it was a good explanation.

[lsatenstein]

HI DJ
I am revisiting zfs/zol again. I have installed Ubuntu with zfs. Are following the software development of zol?

From what I read and from youtube videos, zfs blocksize is 520 bytes. One of my hard drives is the standard type with 512bytes per block, while a newer one is 4096 bytes per block. I am uncertain what my m.2 device formats blocks.

Now my question is about wasted blocks. With 520 bytes per block, would zol/zfs use two of the standard 512blocks to store 512 bytes of data. Is this correct?

For the larger 4096 block, it is not an issue for me. I do not have any of those hardware drives.

My second curiousity is about zol/zfs writing a large file. Zol would checksum the 512byes, and then write out 520. for each 512 bytes a 520byte block is need to land on a new hardware block, which implies that a zfs file, when copied to Linux, consumes twice the diskspace due to the appended checksum?
For my own use, I have not accumulated more than 500megabytes of data. I am really just wondering if those 500megs will consume the one terrabye space.

PS1. Thank you for all your effort at presenting informative videos. I appreciate your analytical approach to the new technologies.
PS 2 I am not enamoured with using btrfs. I/O seems quite a bit slower than ext4, xfs and zfs. I have to benchmarks to do the speed comparisons.

[JD-imwu]

what are your thoughts of Linux vs FreeBSD userland ZFS as you know is native to FreeBSD. I've never made the transition to FreeBSD personally mainly due to using Nvidia as my GPU and I heard of headaches with FreeBSD and hardware so I never tested their stuff out. WOuld you recommend FreeBSD as a workstation over Linux to a low level developer? I use Debian as my base as of now and have to use a VM for testing new software out that is unavailable in the Stable version. I tried Fedora for more updated packages but they crashed me multiple times on updates due to proprietary vid cards and also I found many prior versions glitchy at best. I thought FreeBSD might be a great choice but never had the drive and spared the time to make the switch over or even try it out in a VM due to such a busy schedule lately dealing with outside of computers...

[hamesparde]

You should have showed it on BSD :). I'm typing this on a computer running FreeBSD and I'm going to be using ZFS on another computer running NetBSD.