Solaris + ZFS = The Perfect Home File/Media Server

For a while I’ve been wanting to build the perfect home file server.  When I last lived with a couple good friends, we had a FreeBSD box setup for serving content.  Now keep in mind this was back in 2002, so the whopping 300GB we had online was fairly sizable.  The disks were configured in a RAID 1 array, so as to ensure we wouldn’t lose data if we lost a drive.

Well, we did.  But that was because one drive died and we (ok, I) didn’t replace it before the second one in the mirror also died.  Since then, I’ve been keeping my data on an external 1TB Western Digital drive, all the while being worried that if anything happened to the drive, I would again lose everything.  I needed a file server to maintain all my various data, and it needed to be both flexible and powerful.  Those cheapo NAS devices don’t offer the level of control and access that I wanted.

My requirements for building a server were:

• Secure – I want to be able to tweak and control access at every level
• Able to easily add to and extend existing storage
• Very fast network performance.  I use and love my Popcorn Hour to watch video in HD, so the server should be able to stream 1080p while doing any number of other tasks without issue
• Ability to serve all different types of hosts: NFS, AFP, SMB, SCP/SFTP
• Extremely fault-tolerant.  I’d like to be able to lose two disks and still be ok
• Flexible. I do a number of other things/experiments, and I’d like to be able to use it for more than just serving files.
• Inexpensive.  Self-explanatory, I’m a poor student.

The storage flexibility is something that has been around for ages on expensive SAN solutions from people like EMC. I’ve worked pretty extensively with Solaris over the years, and have been closely following OpenSolaris over the past few years.  Solaris is all of these things, and with ZFS as the filesystem of choice, that same flexibility can be had for a much cheaper price.  ZFS is just the coolest thing to happen to data storage in ages. Read more about ZFS to see just how cool it is. It brings the ‘I,’ which stands for ‘Inexpensive,’ back into RAID.

I ordered the following parts to come up with a fast, fault-tolerant, flexible server:

For a total of $1,334.85, pre-tax/shipping.  Which I think is pretty inexpensive for what you’re getting.  A few notes:

ECC memory.  You don’t need to use ECC memory, but it’s really a bad idea to run ZFS without ECC. ECC memory helps to prevent any internal corruption, and since data integrity is important, it’s worth it. You’re spending all this money to build a file server with redundancy of disks to help prevent data loss, why wouldn’t you build redundancy into memory as well? So don’t run ZFS without ECC.

You’ll see there’s no case on that list.  I reused an old ATX case I had around, which works fine with the microATX Super Micro motherboard.

The motherboard comes with six SATA cables, you may want to order a few extra to have on hand in case you don’t already have some lying around. The power supply has eight SATA power connections, all of which will be consumed by the drives in this setup. So if you ever intend to connect additional drives, you’ll have to get a converter to use the Molex connectors to power them.

The power supply.  Make sure you have enough juice from whatever power supply you use to power all the drives.  Do a little research if you’re unsure, as you don’t want to have a system with drives that aren’t getting enough power.

The extra SATA card.  The mobo only has six SATA ports on it, and the system has eight drives.  This was the cheapest decent SATA II card I could find, with the downside being that I’ll have to replace it if I want to add more drives in the future.

Once it’s all built, download the latest OpenSolaris ISO and follow the install guide. I won’t go over it here, as it’s very straight forward. One thing you should note is that when Solaris installs, it will require you to select an entire disk for the OS to use.  This is where you pick the smaller 320GB drive and let it do its thing.

Once it reboots, it’s time to setup a big ZFS file system from the other drives. You have to decide whether you want to use RAID-Z or RAID-Z2.  The difference between the two is that RAID-Z uses one disk for parity, and RAID-Z2 uses two disks for parity.  If you go with RAID-Z, you’ll have more usable space at the cost of less fault-tolerance: lose more than one disk and you’re toast.  RAID-Z2 allows you to still retain data if you lose two disks, at the cost of losing one more disk’s worth of storage.  After my previous difficulties, and considering disks are so cheap these days, I opted to go with RAID-Z2. Keep in mind that with ZFS if your file system is ever running out of space, you can just toss in another disk, add it to the pool, and you’re done.

CLARIFICATION (updated 5/19/2009): Thank you to Steve and Grant, who correctly pointed out in the comments that this last statement about simply tossing in another disk does not apply to RAID-Z and RAID-Z2 pools. Because of how ZFS is built, it is currently not possible to just extend a RAID-Z/2 set by adding a single drive. Although there has been discussion of how this could be implemented, it currently hasn’t been done. The same applies to shrinking RAID-Z stripes. Steve and Grant are correct, if you wish to maintain the redundancy provided by RAID-Z/2, you must add another set of drives to the pool, thus requiring at bare minimum two drives for RAID-Z or three for RAID-Z2.

So lets get to actually creating that ZFS file system, with the examples using RAID-Z2.

First, identify the disks in the system:

.oO(root@st0wlaris ~) format
Searching for disks...done
 
AVAILABLE DISK SELECTIONS:
       0. c3t0d0 <DEFAULT cyl 60798 alt 2 hd 255 sec 126>
          /pci@0,0/pci8086,2940@1c/pci197b,2363@0/disk@0,0
       1. c3t1d0 <DEFAULT cyl 60798 alt 2 hd 255 sec 126>
          /pci@0,0/pci8086,2940@1c/pci197b,2363@0/disk@1,0
       2. c4d1 <DEFAULT cyl 60798 alt 2 hd 255 sec 126>
          /pci@0,0/pci-ide@1f,2/ide@0/cmdk@1,0
       3. c5d0 <DEFAULT cyl 60798 alt 2 hd 255 sec 126>
          /pci@0,0/pci-ide@1f,2/ide@1/cmdk@0,0
       4. c5d1 <DEFAULT cyl 60798 alt 2 hd 255 sec 126>
          /pci@0,0/pci-ide@1f,2/ide@1/cmdk@1,0
       5. c6d0 <DEFAULT cyl 38910 alt 2 hd 255 sec 63>
          /pci@0,0/pci-ide@1f,5/ide@0/cmdk@0,0
       6. c7d0 <DEFAULT cyl 60798 alt 2 hd 255 sec 126>
          /pci@0,0/pci-ide@1f,5/ide@1/cmdk@0,0
Specify disk (enter its number): ^C
.oO(root@st0wlaris ~)

Since I know c6d0 is the only disk currently in use (it’s the only differently-sized disk), I know all the others will be used in my raidz2 setup. But to be sure, first verify the existing pool’s contents:

.oO(root@st0wlaris ~) zpool status
pool: rpool
state: ONLINE
scrub: none requested
config:
 
       NAME        STATE     READ WRITE CKSUM
       rpool       ONLINE       0     0     0
         c6d0s0    ONLINE       0     0     0
 
errors: No known data errors

Yep, c6d0 is in use. So we’ll use all the other disks. So first, create the storage pool:

.oO(root@st0wlaris ~) zpool create zpool raidz2 c3t0d0 c3t1d0 c4d1 c5d0 c5d1 c7d0
.oO(root@st0wlaris ~) zpool status
pool: rpool
state: ONLINE
scrub: none requested
config:
 
       NAME        STATE     READ WRITE CKSUM
       rpool       ONLINE       0     0     0
         c6d0s0    ONLINE       0     0     0
 
errors: No known data errors
 
pool: zpool
state: ONLINE
scrub: none requested
config:
 
       NAME        STATE     READ WRITE CKSUM
       zpool       ONLINE       0     0     0
         raidz2    ONLINE       0     0     0
           c3t0d0  ONLINE       0     0     0
           c3t1d0  ONLINE       0     0     0
           c4d1    ONLINE       0     0     0
           c5d0    ONLINE       0     0     0
           c5d1    ONLINE       0     0     0
           c7d0    ONLINE       0     0     0
 
errors: No known data errors
.oO(root@st0wlaris ~)

Identify the available pool and file system space.

.oO(root@st0wlaris ~) zpool list
NAME    SIZE   USED  AVAIL    CAP  HEALTH  ALTROOT
rpool   298G  7.33G   291G     2%  ONLINE  -
zpool  5.44T   216K  5.44T     0%  ONLINE  -
.oO(root@st0wlaris ~) zfs list
NAME                     USED  AVAIL  REFER  MOUNTPOINT
rpool                   11.3G   282G    72K  /rpool
rpool/ROOT              3.31G   282G    18K  legacy
rpool/ROOT/opensolaris  3.31G   282G  3.24G  /
rpool/dump              4.00G   282G  4.00G  -
rpool/export            23.2M   282G    19K  /export
rpool/export/home       23.1M   282G    19K  /export/home
rpool/export/home/st0w  23.1M   282G  23.1M  /export/home/st0w
rpool/swap              4.00G   286G    16K  -
zpool                    120K  3.56T  36.0K  /zpool

You can see here the disparity between the first and second commands on zpool. 5.44TB of space, but you wind up with 3.56TB of actual usable space. This is expected, as two of the six 1TB drives have been effectively sacrificed to parity.

Next just create any file systems you want, and you’re done. I know. Criminally easy, isn’t it?

.oO(root@st0wlaris ~) zfs create zpool/media
.oO(root@st0wlaris ~) zfs create zpool/software

And now make them available read-only via NFS:

.oO(root@st0wlaris ~) zfs set sharenfs=on zpool/media
.oO(root@st0wlaris ~) zfs set sharenfs="ro" zpool/media

Each file system shares the same pool of storage as every other file system living within the same pool. But the big advantage is that each file system behaves and can be controlled like a fully independent file system. So you can set quotas, different rules, whatever you so desire.

There is a lot you can do with ZFS, and this just deals with getting it setup in the first place. I’ll be posting more as I do different things with the server.

_{»crosslinked«}