When performing a modal analysis, you can get a lot of modes with low or near zero mass participation. The results is a lot of garbage data. When moving on to a random vib or other vib analysis whose input data is from the modal run, the inputs are overwhelming to the solution.
The suggestion is to put a threshold parameter of "minimum reported mass participation" in the model that will delete the results under that threshold, thereby delivering a clean input set to the random vib analyses.