Many approximation algorithms for NP-hard problems are based on the following scheme: relax the constraints and set up a linear program that yields an optimal fractional solution for the problem at hand; and now "somehow" round this to a feasible integer solution of "low" cost (relative to the cost of the fractional optimal LP solution). How do we achieve this rounding step? How do we prove that the solution obtained, even though sub-optimal is "reasonable"? In this talk we will outline one such algorithm by Shmoys, Tardos and Aardal (STOC 97) for a basic facility location problem and discuss improvements by Guha and Khuller (SODA 98). Both algorithms run in polynomial time, and guarantee solutions within small constant factors of the optimal solution.