What have you tried and where are you stuck?

For normal distributions, there are famous results about what proportion of datapoints usually lie within certain multiples of a standard deviation from the dataset's mean. This looks like a very simple application of those results.