• "Beefing up" a single server is not good enough
• We need to run our service on multiple machines
• If one machine fails, stop sending it traffic and send to others
• Use machines in different locations (cabinets and even data centers)
• This can get pricey and complex
• Virtual machines and the cloud make this much more possible
• Everything can still break. You need to design systems that can fail

7 / 18