Idle and oversized servers are costing you money. You’re not alone in this. Most organizations waste an estimated 29% of their cloud spend. With a few tweaks, you can reduce infrastructure costs without slowing anything down. But this only works when you cut waste first. Here’s how:
Start With a Baseline
Before changing anything or decommissioning idle servers, list every physical server, VM, and cloud instance, along with its owner and its dependencies. Then, collect CPU, memory, disk IO, and network data for the next month using Prometheus, sar, or your cloud provider’s monitoring.
Look at p95, not averages. A server that averages 15% CPU can still hit 90% during nightly backups. Once you have that data, you’ll be confident in shutting down Zombie Servers that nobody would miss.
Shut Down Idle and Zombie Servers
You can’t let servers run for 6 months without them doing any useful computing. To find these zombie computers, flag servers with near-zero CPU and network activity from your baseline.
If no one claims a server within two weeks, power it off. Dev and test servers rarely need to run at night or on weekends, either, so schedule them off outside working hours.
Rightsize and Consolidate What’s Left
There isn’t a universal way to “rightsize.” As with all things in tech, the number depends on your current needs. For example, a company with 5,000 employees might need 110 network switches on paper. But building layouts and what each building did needed close to 500.
Monitor servers’ p95 usage and account for 20%-30% headroom to cover traffic spikes and normal growth. You can resize Cloud servers and VPS plans in minutes. There’s no reason to pay for traffic that isn’t there yet.
A physical server stays in service for 3 to 5 years, so buy for the growth you expect over that time and leave empty memory slots and drive bays for later. Before you add resources to a slow server, find the bottleneck.
File and database servers often run out of storage IOPS before they run out of CPU. Consolidate servers that aren’t often in use. Move them onto virtualization hosts you already run.
Move Rarely Used Data to Cheaper Storage
You’ll rarely open backups, old logs, and finished project files—but you might need them in the future. The issue is that they often sit on the same SSDs as your production databases. Storage tiering puts each kind of data on the cheapest tier that still meets its needs:
- Production databases and VM disks belong on SSDs.
- File shares and old logs can live on regular hard drives or standard cloud storage.
- Old backups and compliance records can be archived to storage services such as Amazon S3 Glacier.
Set lifecycle policies to move data to a cheaper tier after a specified period. Delete orphaned disks and old snapshots you don’t need.
Keep anything your apps read often on fast storage, since cold storage can take hours to restore. Test a restore from each tier before you rely on it.
Buy the Capacity You Need for Less
AWS Savings Plans cost up to 72% less than On-Demand pricing. But you should commit to these savings plans only after you have removed waste and established your actual baseline usage.
For small, steady workloads, a VPS server provider with flat monthly pricing can cost less than a comparable cloud instance. AWS Spot Instances cost up to 90% less than On-Demand prices.
The catch is that AWS can take a Spot Instance back with 2 minutes’ notice. Spot Instances work best for jobs that can restart, like nightly reports or automated code tests.
If you replace aging on-prem servers, you also reduce running costs. Newer servers are more efficient. They can do the same work on fewer machines.
Make the Savings Stick
Server costs tend to creep back up after a cleanup. Teams spin up “temporary” servers that never get shut down, and rightsized servers get resized upward. Here are three habits that keep server costs from creeping back:
- Tags: Label every server with an owner, app, and environment
- Budgets and alerts: Catch spending spikes the week they happen
- Quarterly reviews: Re-run your baseline and look for new zombie servers
