Just finished working with a client who had setup a straightforward AWS account with several web servers, a database server … the usual. He had run all the cost analyzers for server size, instance types and reserved versus on demand contracts. His bill should have been about $1500/month, but when he checked halfway through the month on his costs, he was on track to spend 10k for a month.
The clue that helped me solve me the problem is the EC2 Other category in analyzing the bills. This category includes various things related to EC2, like load balancers, disk space and i/o. When I checked the configuration of the disks, the system admin had multiple 900 GB drives with provisioned IOPS in the 20k per second range. Nothing in their hosting environment needed that kind of I/O.
The fix was as easy as the problem. I simply converted the disks in the volume manager from provisioned IOPS to SSD with default I/O settings. No issues, and the bill shrank to the expected size.
This is a bit of a gotcha in AWS. When you’re provisioning disks for servers choosing between provisioned IOPS and SSD is as simple as choosing a value from a dropdown, but there’s no warning that your choice may incur a $300 a day charge.
Leave a Comment