A method of operating a supercomputer having N computing elements each
connected to a fast communications link is disclosed, the method
comprising the steps of: operating the supercomputer to perform a
computing operation; upon failure of a fast communications link
transferring state from a computing element which, as a result of the
fast communications link failure, is no longer able to communicate, to a
spare computing element not previously engaged in the computing
operation, and continuing the computing operation with the spare
computing element, wherein the number of redundant elements M is chosen
to satisfy the expression B.sub.M[N, (1-P.sup.T)]>S where S is a
desired probability of successful completion of the computing operation
within a time T and P is the probability of successful operation per unit
time of a fast communications link.