> The reason for this is that you want to be able to saturate your > storage subsystem, and that means keeping all spindles working at all > times and efficiently. This is accomplished by ensuring you are able > to sustain a sufficient queue depth (number of outstanding commands) > on each device. This in turn, in the case of a RAID0, means > multiplying the target maximum queue depth with the number of drives.
(But this is all predicated on the operating system actually letting the I/O requests pass through to the device, which is why I replied about choosing the deadline or noop scheduler instead of cfq.) -- / Peter Schuller
