http://www.sohoconsult.ch/raid/images/raid10.gif
Draw a vertical line down the middle of this illustration,
right through the "m" in "Performance":
On the left are mirrored data: a "chunk" of data named "A"
gets written to both spindles; then, the next "chunk" named "B"
gets written to both spindles; and so on, until the entire
file is written to both, resulting in two (2) copies of the file:
one copy on one spindle and a second copy on the other spindle.
On the right are striped data: a "chunk" of data named "A"
gets written to one spindle; then, the next "chunk" of data named "B"
gets written to the other spindle; then, the next "chunk" named "C"
gets written to the same spindle as the "A" chunk; then,
the next "chunk" named "D" gets written to the same spindle
as the "B" chunk; and so on.
Mirroring (RAID 1) can only go as fast as a single spindle;
striping (RAID 0) goes faster than a single spindle, but
never reaches twice as fast because of rotational
latencies, armature seeks, and other sources of
overhead such as controller efficiency (or lack thereof).
Even with solid-state disks, which eliminate rotational latency
and recording head movements, "scaling" is not linear when
the number of drives in a RAID 0 array is increased from
2 to 4 to 8 devices. Thus, a RAID 0 array with 2 devices
will generally NOT operate twice as fast as a single device,
but at a rate significantly less than twice as fast.
Lastly, a RAID 10 set is merely 2 x RAID 0 arrays that are also mirrored
i.e. both RAID 0 arrays have identical copies of exactly the same data.
MRFS