I thought I would write something about my experiences of zfs as a system administrator, as I have been running it on my home server for a long time now (over a year). Other people will tell you about all about what is under the hood and how the checksums and whatnot works, but I'll give you my "user experience".

Before I started to use zfs on my Shuttle SB51G I had all my data (home directories, Solaris workspaces, etc) stuck in one big UFS filesystem monuted on /export. From where I loopback mounted directories to where I wanted them, e.g. /export/local to /usr/local.

One of the things that annoyed the <beep> out of me with this approach, was that I constantly had to run du to figure out what was hogging the disk. With zfs this problem is gone, as filesystems are cheap and easy to create, I stick everything in its own filesystem, and I can easily see where the culprits are. I'll show how to look for this below.

Since I started out with zfs a long time ago, I had to bfu to get access to zfs, and then I could to start out by backing up all my data (using tar). Once I had my backups in a safe location I undid the lvm mirroring on my /export partition and added the slices to a storage pool:

# zpool create p1 mirror c0t0d0s7 c0t1d0s7

That's it! I just configured a mirrored pool, no more creating meta databases, mucking around with md.cf and running metainit. When you use lvm you have to make an estimate on how much space you will need in each filesystem, which can be very hard to do accurately. Now that we have storage pools you don't have to do that any more, as all filesystems you create will dynamicaly use space from that pool. To prevent a single filesystem from taking up all available space in the pool, you can create quotas for filesystems, but more about that later.

Now I could take a look at the pool I just created:

# zpool list
NAME                    SIZE    USED   AVAIL    CAP  HEALTH     ALTROOT
p1                    426.3G      0K  426.3G     0%  ONLINE     -

And if I want to see exactly how the pool is laid out:

# zpool status -v
  pool: p1
 state: ONLINE
config:

        NAME                  STATUS
        mirror                ONLINE
          c0t0d0s7            ONLINE
          c0t1d0s7            ONLINE

Having a storage pool isn't enough, I needed some filesystems too, so I could restore my data:

# zfs create p1/home
# zfs create p1/home/martin
# zfs create p1/home/ftp
# zfs create p1/local

Now I have one container and three filesystems, which by default are mounted under /zfs/p1/home and /zfs/p1/local which isn't where I want them to be, so I have to modify that:

# zfs set mountpoint=/home p1/home
# zfs set mountpoint=/usr/local p1/local

and now they are and will always be mounted on those locations.

Two nice features of zfs are the quotas and reservations. On my anonymous ftp account I don't want people to be able to upload too much junk, so I place a quota on that filesystem:

# zfs set quota=10g p1/home/ftp

And since I am greedy, I want to make sure I always have plenty of space in my home directory, so I give myself a nice little reservation:

# zfs set reservation=50g p1/home/martin

There is another cool feature in zfs which I use on all my Solaris source code workspaces, namely compression. I start out by creating a container to hold all my workspaces:

# zfs create p1/ws
# zfs set mountpoint=/ws p1/ws
# zfs set compression=on p1/ws

Now every filesystem I create under p1/ws will have compression enabled:

# zfs create p1/ws/onnv

so I won't have to go in and enable it for each new filesystem I create.

To verify that the settings are inherited from the p1 pool and p1/ws container I use:

# zfs get -a p1/ws/onnv
PROPERTY      VALUE                      SOURCE
type          fs                         -
used          5.5G                       -
available     385G                       -
creation      Mon May 30  2:37 2005      -
referenced    27.5K                      -
ratio         72.10x                     -
mounted       yes                        -
recordsize    128K                       default
quota         none                       default
reservation   none                       default
mountpoint    /ws/onnv                   inherited from p1/ws
sharenfs      rw                         inherited from p1
checksum      on                         default
compression   on                         inherited from p1/ws
atime         on                         default
devices       on                         default
exec          on                         default
readonly      off                        default
setuid        on                         default
zoned         off                        default

Now when everything was setup the way I wanted I just had to restore my data into the right locations and I was ready to start using zfs for real.

Once all the data was resored, I could easily see what was taking up the most space:

# zfs list
NAME                   TYPE   USED  AVAIL  MOUNTED  MOUNTPOINT
p1                      ctr  41.3G   385G        -  /zfs
p1/home                 ctr  35.8G   385G        -  /home
p1/home/ftp              fs     2M    10G      yes  /home/ftp
p1/home/martin           fs  35.8G   385G      yes  /home/martin
p1/local                 fs     6G   385G      yes  /usr/local
p1/ws                   ctr   5.5G   385G        -  /ws
p1/ws/onnv               fs   5.5G   385G      yes  /ws/onnv

The bottom line is: this is the only filesystem I use now, apart from UFS on my root partition (which is going to change). I am happy as a clam with zfs. It has made filesystem administration so much easier!

[Technorati Tags: OpenSolaris Solaris ZFS]