The real-time convolution of a sound signal with measured or computed room-impulse responses poses severe problems as regards computation load, system latency and memory requirements. Some key features of frequency-domain convolution allow to solve the dilemma between latency and computational complexity. Extrapolation from nowadays implementations on off-the-shelf computers allows to sketch future software solutions. In addition the properties in structure and dynamic of the room-impulse responses match ideal with those of the low-latency convolution algorithm for real-time spatial sound processing.