More attention should be paid on system noise for large-scale parallel application, although the system noise has little impact on one process. One quantitative analysis method for system noise's impact named FWQ-MPI is presented. Four quantitative indicators are given: the proportion of the amount of noise, the proportion of noise impact, the actual/ideal ratio of communication to calculation time. Three iterative methods are selected as the research objects and the micro benchmarks run on a MPP machine with 512 Double six-core nodes. The test results show the impact mechanism of system noise on parallel program performance, and also show several characterizations of system noise as follows. 1)The proportion of the amount of noise is relatively small, accounting for the entire computing time 2% to 6%; 2)But the system noise has much impact( when the parallel scale is 1024,2048 and 4096, the proportion of noise impact is about 30% to 70%); 3)The impact of system noise will be increased when the parallel scale increases, will be decreased when the amount of calculation increases; 4)The impact of system noise is mainly reflected by the “actual ratio of communication to calculation time” is far less than the “ideal ratio of communication to calculation time”.