Originally Posted by CrazyOne
One thing I was told over and over by cautious types when I was thinking of doing some RAID at work is that you don't trust critical data to a striped array. So I'll repeat that here and let any less cautious types refute it if they like.
It's not a matter of the cautious types: if you lose a spindle in a striped (RAID-0) configuration, you lose the overlaid filesystem. Period.
A brief primer for all who might be interested (spindle, in this case, represents one harddrive):
Let me begin by saying that you could look at your RAID-level choices thusly: fast, cheap, reliable--pick two.
RAID-0=striped with no parity or redudency. Makes for one very large logical disk, whose read and write speeds are typically very, very fast (or, somewhat higher than a single spindle).
RAID-1=mirrored data. If you have 1 drive's worth of disk, a RAID-1 array will need to consist of two drives. Read operations on expensive RAID controllers can be faster than a single spindle (most home and desktop implementations are not). Writes are equal or slower than a single drive, especially with one disk controller (you have to commit the write to two disks).
RAID-2=stripes the data at the bit level (not the disk block level). Nobody does this anymore.
RAID-3=Byte striping with a seperate (dedicated) parity disk. Nobody does this anymore.
RAID-4=Block striping with a seperate (dedicated) parity disk. Writes are slow, reads can be fast. Nobody but Netapp does this anymore.
RAID-5=Block striping with parity and data striped across all disks in the array. Reads (particularly random ones) are fast-ish, writes can be slower than a single spindle depending upon caching and controllers. This RAID level is useful because if you blow one disk in the array, you don't lose the data--most RAID-5 arrays can operate in a "degraded" mode on array-1 disks. Most will also automatically rebuild the data/parity on the blown disk, by reading/calculating it from the remaining disks. In this sense, it's very interesting because (under certain circumstances) it's both fast and redundant.
These are the basic RAID levels. You can also do things like RAID-0+1 (a mirrored stripe), RAID 1+0 (a stripe of mirrors) and some other things. Some companies, notably EMC, have their "own" RAID implementations (EMCs is called RAID-S, where they simply have a novel way of calculating and storing parity data).
So, now that we are confused.....
For the OP: If you want a safety net, get yourself a relatively cheap RAID controller, throw two identical drives on it, and move on. Or, buy a Lacie, but
make sure it's not
merely striped internally (see RAID-0). If your G-V is running OS-X (or whatever animal Apple is using now), it's not entirely out of the realm that it can do a software mirror on external firewire drives (I'll leave that one alone--I'm not familiar with the "apple-ized" BSD
).