A computer system includes a plurality of field replaceable units, each
having volatile memory and at least one CPU. The FRUs communicate with
each other via centralized logic. A RAID data fault tolerance technique
is applied to the system so that an FRU can be lost or removed without
loss of its data. An exclusive OR engine is included in the centralized
logic or distributed among the FRUs. The RAID logic can restripe itself
upon removal or addition of a FRU.