What performance metrics do you find most critical when monitoring large-scale applications? How would you implement monitoring for those metrics?