A CPU design technique that periodically "grabs" machine cycles from the main processor usually by some peripheral control unit, such as a DMA (direct memory access) device. In this way, processing and peripheral operations can be performed concurrently or with some degree of overlap. See also
peer-to-peer computing.