Hacker News Reader

Comment by jrochkind1 | Hacker News Reader

While it's more frustrating to have no control over the reliability, I suspect that most companies could not achieve the level of reliability from AWS by self-hosting, at anywhere close to the cost (inc. stafftime). Nothing is 100%, and getting as close to 100% as you can gets exponentially more expensive the closer you get.

Of course, this could change if AWS's reliability trends downward. But it's never going to be 100%, and neither is anything you do yourself.

View on HN

I don't think this is true. US-East's incident history over the past 3 years is terrible. If you had a completely non-redundant set of rented servers sitting in Softlayer or Rackspace's data center over the same period, even factoring in expected hardware failures over the time period (replacing a hard drive or two, probably), you'd have been available more than even a multi-AZ deployment with failover in AWS US-East.

I've got heavily used servers at Softlayer that haven't had more than a few minutes of downtime in 4+ years, and have never been rebooted. During that same time, I ran a site out of US-East for two years (EC2 + EBS + Multi-AZ RDS + ELB), and had more downtime and spent more admin time working around significant issues with Amazon. Amazon neither saved me time nor money, nor did it provide better availability.

wonder if anyone's been keeping and publishing statistics on exactly how often AWS US East has been down. Perhaps it's more than I thought. Of course, most failures haven't been effecting _all_ of US East, so maybe what would be interesting to look at is just _some_ customer's data on exactly how much downtime they had in the past 3 years or whatever. Without knowing what the uptime/downtime was, hard to say how terrible it was, or how it compares to other options.

I have a server for a client in our colo, that has been up for 1753 days. (Client doesn't want to move to new hardware, etc. otherwise we would have moved them long ago).

During that time, there was maybe network outage (at 2AM mostly) of perhaps 2 hours.

Complexity reduces reliability.

Claiming AWS' uptime approaches that of fairly standard colocation in a decent facility, is laughable.

Reply on news.ycombinator.com