"Gmail, and we thought it best to simply share our reliability metrics, which we measure as average uptime per user based on server-side error rates." If you a

"Gmail, and we thought it best to simply share our reliability metrics, which we measure as average uptime per user based on server-side error rates." If you are just using error rates from the requests that make it to the application then it seems this measure of reliability doesn't account for a lot of infrastructure at Google (ie, load balancers, switches, etc). Service interruptions and downtime with infrastructure like this should definitely be accounted for by a measure of reliability for Gmail since the end-user needs all of it to work to use the app. One solution could be factoring in some sort of availability monitor at the border of Google's networks that polls/uses the Gmail service like an end-user.