We present a new RNS modular multiplication for very large operands. T
he algorithm is based on Montgomery's method adapted to mixed radix, a
nd is performed using a Residue Number System. By choosing the moduli
of the RNS system reasonably large and implementing the system on a ri
ng of fairly simple processors, an effect corresponding to a redundant
high-radix implementation is achieved. The algorithm can be implement
ed to run in O(n) time on O(n) processors. where n is the number of mo
duli in the RNS system, and the unit of time is a simple residue opera
tion, possibly by table look-up. Two different implementations are pro
posed, one based on processors attached to a broadcast bus, another on
an oriented ring structure.