[tritchey]

We recently launched a simple website monitoring app for the iPhone where we ping sites from our servers, and send push notifications when we detect something is wrong (a poor man's pagerduty or pingdom where you do everything from your phone and can subscribe using IAP).

Probably the biggest lesson we've learned from collecting data on latency and timeouts from our backend is that there are a LOT of very poor hosts out there. Trying to strike a balance between intermittent failures, and actual "Your site is down, you need to do something" is pretty difficult, unless you are willing to simply wait long enough. We'll regularly see timeouts from places like Tumblr and Github Pages--they never go down, but things can get spotty. We can also tell when certain hosting providers are doing their weekly backups (or something - just guessing) because things will just go to pot at the same time every weekend for about an hour.

I can second the S3/CloudFront recommendation. Pretty rock solid (modulo downtime in specific regions).

For anyone curious about the app: http://vigil-app.com