How would you perform k-Means on very large datasets

How would you perform k-Means on very large datasets?

What is your concern about performing K-Means on very large datasets?