‘Five Nines’ refers to the five nines in 99.999% available that is often synonymous with highly available. Does every highly available service require five nines? Not by a long shot. Yet the general state of the practice is to chase after this typically unrealistic goal almost blindly in many cases, often leading to unnecessarily high costs in both operational and development resources. Even less aggressive availability goals are often over-specified compared to true business drivers.
This talk will cover: * the history of ‘five nines’ * common reasons why many organizations often inadvertently over-specify availability requirements * the costs of such over-specification * how service agility is negatively affected * examples of highly available systems with reasonable availability requirements * techniques on how to avoid over-specification based on Site Reliability Engineering principles * ways to spend your Error Budget (once you have one) most effectively
Applying these techniques should result in a more cost-effective service that keeps end users and management happy, and fewer alerts to the on-call DevOps engineer.