At this year's Cloud Connect event in Santa Clara, Calif., experts cautioned that careless provisioning could lead to major surprises on monthly bills.
The problem first gained attention around a year ago, when InformationWeek's Beth Stackpole reported on the launch of usage-monitoring services like Cloudyn, Cloudability, and CloudCruiser. At the time, I thought the names were pretty unimaginative and essentially placeholders pending acquisition by one of the big CSPs.
But here we are a year later, and apparently, many cloud customers are still letting overruns and mistakes ruin their budgets. At Cloud Connect, InformationWeek's Charles Babcock attended a workshop led by executives from Cloudyn and Cloudability called "Optimizing Your Cloud Spend: How to Do It Right." The presenters shared some surprising findings that emerged from reviewing thousands of customer usage patterns.
Cloudyn CEO Sharon Wagner shared that some 80 percent of public cloud server instances are underutilized while 15 percent are overutilized, leading to latency and poor service. Average CPU usage is 17 percent, which is better than the stats for physical servers but still could be better.
He recommends paying upfront for reserved instances instead of paying a premium for on-demand resources, but only 19 percent of his customers do.
Mat Ellis, CEO of Cloudability, shared a few "horror stories" of cost overruns created by forgotten instances. In one example, he racked up a charge of $1,600 for resources he never used, simply because they were deployed in Amazon's west coast datacenter, and he assumed all of his resources were on the east coast.
In another example, a software developer firm set its budget for the business week, not realizing that employees kept their instances running on weekends and holidays, leading to an extra $350,000 over budget.
To me, all of this boils down to infrastructure assessment and a little advance planning, but that's obviously not as simple as it sounds. Ellis laid out five best-practices:
- Use the CSP's reporting tools or a service like Cloudyn or Cloudability to track usage.
- Check each month to make sure employees are using the resources they ordered.
- Consolidate company-wide usage into one bill, both to prevent rogue departments from striking out on their own and to negotiate better prices.
- Try to trim instance sizes and uptime to fit your actual usage needs.
- Tag every resource with the group and project using it.
Members, have you encountered any nasty cloud bill surprises? Share your experiences with us here.