But read the fine print and then do the maths. What Amazon is promising is a product that won’t crash more than once a week. Enterprise customers: you’d be insane to transfer mission-critical IT applications like email or ERP onto EC2 if you don’t believe Amazon can deliver more reliability than that.
Why? Because Amazon says it measures “uptime” in five-minute increments. So when the company touts 99.95% uptime, that means it promises not to miss more than one every 2000 five-minute blocks of compute time. Which means it considers a failure every 166.7 hours OK. Or about once a week.
For some applications, that’s probably OK. Outsource the CPU heavy lifting on a lot of web applications to Amazon, and during the expected once-a-week outage, annoyed websurfers will probably click “refresh” on their browsers a few times and service will be restored. But what about email? Is it an acceptable standard to know that every week, on average there will be at least one moment when your organisation’s mailservers bounce?
Or worse — ERP. Systems like SAP’s (SAP) or Oracle’s (ORCL) payroll and general ledger programs rely on a block of time every night — called the “batch” — when systems churn though a series of sequential programs all night long to move a company’s books forward one day. A single problem anywhere in the line can cause entire IT staffs to be woken at midnight and leave companies paralysed in the morning until the ERP system catches up on lost time.
And only one outage every seven days is the most Amazon will guarantee. What if Amazon achieves less? The company says if you experience outages more than the advertised once-a-week, and if you can document it to their standards, they’ll issue a credit equal to 10% of your bill you can apply towards future EC2 service.
Disclosure: Our parent company shares investors and office space with 10gen, a cloud computing startup.