Finding cloud waste you can prove

Cloud bills rarely grow because of one bad decision. They grow because of a long list of small things that were each reasonable at the time: a disk kept after its server was deleted, a test environment nobody switched off, a load balancer in front of nothing.

Each item is too small to chase on its own. Together they are often the easiest saving a team can make, because removing them changes nothing about how the product works.

The usual suspects

These turn up in almost every account that has been running for more than a year:

• Disks with no server. Deleting a virtual machine does not always delete its disks. They keep costing the same as when they were attached.
• Old snapshots. Backups from machines that no longer exist, kept with no expiry.
• Idle load balancers and gateways. Charged by the hour whether or not any traffic passes through. NAT gateways in forgotten networks are a common one.
• Public IP addresses attached to nothing.
• Stopped machines. The compute charge stops. The disk charge does not.
• Oversized machines. Sized for a launch day that has passed.
• Non-production environments running all night. A development environment used forty hours a week is billed for a hundred and sixty-eight.
• Whole projects nobody remembers. A proof of concept from two years ago, still running.

Proving it is waste

Deleting the wrong thing is worse than paying for it, so each candidate needs evidence. A short checklist:

1. Is it attached to anything? A disk with no server, an address with no interface.
2. Has it done anything recently? Requests, connections or processor use over the last few weeks, not the last few minutes.
3. Who created it, and when? The change record usually tells you, and the person often knows straight away whether it is safe to remove.
4. What does it cost each month? A number makes the decision easy to prioritise, and easy to explain.
5. Can you get it back? Take a snapshot before deleting a disk. Stop a machine for a week before terminating it.

The best cleanups are boring. You remove something, nothing happens, and the bill is smaller next month.

Keeping the list short

A cleanup is a one-off. Waste comes back unless something changes in how resources are created:

• Tag everything with an owner and an environment. Waste with an owner gets cleaned up. Waste with no owner waits for an audit.
• Give temporary things an end date when you create them.
• Switch off non-production compute out of hours on a schedule, with exceptions for what must stay up.
• Set a budget per account and alert before the end of the month, not after the invoice.
• Look regularly. A monthly scan finds a disk a few weeks after it was orphaned. A yearly one finds it after twelve bills.

How Zaysa helps

Zaysa scans AWS, Azure, Google Cloud and DigitalOcean for wasteful resources. Every finding says what it is, why it is waste and what it costs each month. You can remove waste straight from the finding, with the change recorded. Schedules stop idle compute out of hours, with exclusions for anything that must stay up. Monthly budgets per account come with a forecast and alerts by email or Slack.