Embedded computers exist that draw only 2 watts or less for their operation. As a trade of for this, they are slow, often 10 times slower than a desktop computer. They are meant to be on all the time, recording probe information to send to computers that are only on 9-5.
RAID is a system to combine hard drives in various ways to gain various useful effects. The most common ones are mode 0, which has two (or more) hard drives act like one large combined drive, gaining extra space at the cost of reliability, as the entire array is destroyed if one drive fails, mode 1, in which two (or more) drives copy each other, so the information survives the failure of one of them, and mode 5, in which the information is split between at least 2 drives, with a third one keeping a parity checksum. The advantage of mode 5 being that if one uses hot-swappable drives, the array can run 100% of the time. (This being because if one drive fails, the computer can still determine the information with the checksum. One should swap out the defective drive for a new one as soon as possible, as a second failure will cost you the information.) RAID can also be stacked, most popularly 0 and 1 together for the combined advantages of both.
I think we should combine these technologies for an embedded RAID device that specializes in storing information for the rest of the computers on the network, which can be diskless terminals. The CPU of the RAID device doesn't need to be very powerful, but it does need to run all the time. Probably it will still end up using as much power as a desktop computer, since the savings from the low-power CPU are canceled out by the need for many many many hard drives, but one does get bigger and more reliable storage that's infinitely mobile.