TS-7800: Linux RAID NAS for the Home User?

Update-20100328: Seeing as how Marvell are essentially the only ARM licensees who have gigabit ethernet technology, this device would decidedly make a really cool control board for the USRP2 from Ettus Research, Inc. Luckily, I've been doing some hacking with libusrp2 lately ;-)

I've had my TS-7800 for quite some time now, and I haven't really done anything fantastic with it yet, although, my original plans were to use it quite suitably as a NAS. The TS-7800 uses the Marvell Feroceon MV88F5182 CPU, and while lacking vector floating point (VFP) support, it is perfect for use as a storage controller.

This would be used by Erin & myself for fault-tolerant storage of important data.

Now, when I say fault-tolerant, I'm speaking of a redundant array of inexpensive disks (RAID). The TS-7800 does not come with a hardware RAID controller, as they are very expensive and usually only found in bulky server equipment, but any linux appliance is perfectly capable of acting as a software RAID controller.

The TS-7800 only supports two SATA connected drives, which means that if I only include these two devices in my software RAID configuration, the RAID level is limited to RAID 0 ("striping", improves speed, no redundant storage), or RAID 1 ("mirroring", data redundancy, slight speed overhead). Of course, I would choose the second option to have our data stored reliably. In that case, my capacity would be half of the total disk capacity.

If I were only including two SATA connected drives in my configuration, my RAID configuration would be limited. However, if I chose to make use of more devices connected via the TS-7800's USB ports, then I could have a RAID 5 configuration (striped, with parity). Just like a RAID 1 configuration, no data is lost if one disk is damaged. Unlike the RAID 1 configuration, rebuilding the damaged disk means more than simply copying the information from the other disk. Furthermore, if I implemented this and mixed SATA / USB disks, then there would surely be a performance hit when reading from / writing to a USB disk, which would slow down the array as a whole since operations would be striped across all disks in a RAID 5 configuration. RAID 5 also performs poorly in software when small files are being repeatedly updated because of the parity calculation requirement.

I think, in the end, I'll choose to use two 500 GB SATA disks, 3.5" in size just because the price is so relatively low, such as these, and start off with a simple RAID 1 configuration. Later on, if I feel the need to upgrade the capacity of the disks, or the raid array itself, I can hopefully buy some more similar disks and connect two more via USB, in spite of the slowdown that it might cause.