For this scenario, I tested with two motherboards. Primarily I used a Gigabyte GA-970A-DS3, and then briefly I tested hardware migration with my current system based on the MSI 970A-G45 motherboard. Both of these boards use the same AMD SB950 southbridge to provide RAID support.
Ease of Use
Setting up the RAID in the BIOS is pretty simple. Simply enable RAID in the BIOS and then when prompted during boot enter the RAID manager and create a Logical Drive with the RAID type and the disks you want. Once booted into your OS, the drive is presented as a single physical disk. Drivers are still required under Windows.
Moving the drives from one system to another seemed to detect and load the logical drive in the RAID manager perfectly.
Recovery is not intuitive without the AMD RAIDXpert utility. The only apparent way to replace the failed drive is to delete the logical drive array and then build a new one. This is about the most terrifying thing in the world when your data is at stake. And yes, I realize “You should have backups of everything anyway”, but let’s get real. Critical data will be backed up off site, as covered in the server summary.
With the RAIDXpert utility (Windows only), recovery is much simpler and less terrifying. In fact, it is even possible to set up a hot spare, which will be utilized automatically in the event of a failure. I set up a hot spare and then unplugged the drive. Within a very short time after booting, the failure was detected, the hot spare was automatically pulled into the array, and rebuilding was started. Hot spare’s can be dedicated to a specific array when multiple RAID arrays are in operation, or they can be in a global pool for any array that needs to start rebuilding.
Notification
Drive failure was not reported to Windows in any warning or error level events. I consider this a failure. The system will not be rebooted often, and when it is it will most likely be unattended during an automatic update. The server will probably be headless too (no monitor, at least not on), so seeing a warning only flash across the screen momentarily during the POST boot screen is not good enough. I do not want the drive failure to be hidden from me.
After replacing a ‘failed’ drive in my test, I would expect to see some kind of indication that it is rebuilding the missing drive. I saw no indication either by drive access or status messages in the RAID manager or in Windows. I figured since I had less than 1MB of data on the array, it could rebuild quickly. I then pulled a different drive, and the array was completely lost. Disk Manager merely showed an uninitialized drive of 2N capacity, which is what I’d expect, but I should still have my data. I consider this unacceptable. I understand it takes time to rebuild the replaced drive, but no indication of any kind is unacceptable.
Notification with RAIDXpert
Installation of RAIDXpert brings events into Windows. Unfortunately, that’s about all it gives a home user. The utility has a means of setting up e-mail notifications, however AMD falls short (pending response of technical support e-mail: 12/19/2012). The option for sending mail requires an smtp server, but it provides no option for port, encryption, or authentication. Since the default SMTP port 25 is blocked by most ISPs these days, e-mail notification fails without these options.
Update: (November 23, 2013)Since this was my biggest problem with this solution, I did a little research on alternatives. What I have been running successfully for nearly a year is an open source service called E-MailRelay. This application can be set up as a service that opens a standard open SMTP port 25 and then relays out to a secure SMTP server with login. By default it does not allow remote connections, and has several optional authentication and white listing measures if you wish to open it to external traffic.
Performance
Performance was pretty good when using the caching option. Even using direct writethru, performance is better than the Windows Home Server RAID5 implementation. If caching is to be used, the system should have a battery backup system. I consider this a reasonable expectation. Some write speeds even improved once a drive was removed from the array as a “failure”. Presumably, writing to a failed RAID5 array just falls back to straight striping in this implementation.
 RAID 5 – Caching |
 RAID 5 – Caching – 1 drive “failed” |
 RAID 5 – WriteThru |
 RAID 5 – Write Thru – 1 drive “failed” |
 RAID10 – Caching |
 RAID10 – Caching – Down 1 drive |
Single drive (reference)
Verdict
Ease of Use: 4/5
Drive loss limit: 1*
Performance: 5/5
Expansion / Upgrade options: Pending.
Efficiency: N-1 (RAID5); N/2 (RAID1 / RAID10)
Cost: Free with many motherboards.
Viable: Yes
*Sometimes RAID 10 is listed as 2 or half, but that depends entirely on which drives