Break me if you can! How to Test the Ability of Large-Scale, Distributed Software Systems to Cope with Failures?