On the sampling of big mass spectrometry data