Auto-Scaling

1. The Incident

The stateless fleet behind its balancer runs eight servers, sized for the evening peak of 4,000 requests per second. It's a fine fleet. Here are its two bad habits, discovered in the same month.

The 4 AM bill. Average traffic is about 900 requests per second: the evening peak is 4-5× that (the arithmetic lesson's peak rule, behaving exactly as promised). Which means that for most of every day, five or six of the eight servers are doing nothing, at full price