This paper describes a new data acquisition system at JET which uses Texas TMS320C40 parallel digital signal processors and the HELlOS parallel operating system to reduce the large amounts of experimental data produced by fast diagnostics. This unified system features a two level trigger system which performs real-time activity detection together with asynchronous event classification and selection. This provides automated data reduction during an experiment. The system's application to future fusion machines which have almost continuous operation is discussed.