Among other examples he cited, one customer, a software development company, budgeted for everything it used during the work week over the course of a year, not realizing development teams left everything running on weekends and holidays as well. It had a budget of several million dollars for cloud use; the final bill came in for $350,000 beyond what it had budgeted.
Another company was aware of its regular business and IT cloud bill but didn't know its five-member development and operations team also was spending $5,000 a month in the cloud. The result: an unexpected $60,000 bill.
Ellis outlined the following five steps to avoid surprises.
1. Determine Your Usage.
"Don't accept, 'Oh, we don't use all that much,'" he said. Cloud services report usage monthly with the bill, and reports are available on usage as the month progresses for customers who care to ask for them. And, of course, third-party usage tracking services such as Cloudability and Cloudyn can help with graphic and detailed bills.
2. Check Usage.
Check every month to see whether you're using the resources that your employees have initiated. Look for signs of low usage and start asking questions.
3. Seek Out Price Breaks.
Consolidate all employees using a service provider into one bill, instead of getting several. It might qualify you for a volume price break, Ellis advised. And use AWS' reserved instances, whenever possible, for its hourly rate reduction. Only 20% of Cloudability's customers use reserved instances, he noted.
4. Go On A "Diet."
Watch for how your usage varies and try to right size your servers. Employ Amazon's autoscaling to fire up more instances when you need them and decommission them during much of the time when you don't. Meanwhile, you might be able to downsize your 24-hour-a-day server to something more appropriate.
5. Tag Your Resources.
Use tags to label your resources. This lets you know what resources are being used for and seek explanations when usage goes up. Current cloud billing tends to be poorly labeled. When you can see the bill is bigger it can still be hard to see why. Charge back to users, and keep users informed with weekly updates, Ellis said.
A questioner in the audience pointed out that cloud usage was basically undoing the gains inside the data center of virtualization -- consolidating applications on fewer servers. Now users were initiating more servers outside the data center with low usage rates, all over again.
Both Wagner and Ellis conceded that was "somewhat true today, but it's early in cloud usage" and better usage patterns will restore the savings.
Big iron is under relentless pressure to prove its value in an x86 world. We look at five ways to bring it into the distributed data center. Also in the new, all-digital Mainframes In The Age Of The Cloud special issue of InformtionWeek: Five steps to a better wireless LAN. (Free registration required.)