EBS Volume Rightsizing Discipline
Most EBS volumes are oversized. The audit that catches it and the savings that follow.
The audit
Per-EBS volume utilisation: how much storage is actually used versus provisioned? CloudWatch metrics show used bytes and total bytes for filesystems with the CloudWatch agent installed.
Per-EBS IOPS: provisioned versus consumed. gp3 lets you provision IOPS independently of size; gp2 ties IOPS to volume size. Many teams overprovision both.
Compile a list sorted by absolute waste (provisioned minus consumed). Top 10 typically account for 60-80% of optimisation potential.
Right-sizing down
Storage utilisation under 50% sustained for 30 days: candidate for resize down. EBS volumes can be shrunk by snapshotting, creating a smaller volume from snapshot, and replacing.
Reality: shrinking is operationally risky. Most teams replace at instance refresh rather than online. Plan the resize during the next deploy or maintenance window.
IOPS utilisation under provisioned: drop the IOPS provision. gp3 with 3000 baseline IOPS is the default; only pay for higher when consumed metrics justify.
Upgrade gp2 to gp3
gp3 is universally better than gp2 for new workloads. Same storage, higher baseline performance, lower cost per GB.
Migration is a single API call (modify-volume) per existing volume. Online; no downtime. Performance often improves; cost drops 20%.
Why teams haven't done this: it requires explicit action across the fleet. Do it: it's free money.
Typical savings
First-pass audits find 30-50% of EBS spend reducible without performance loss. Mostly oversized provisions and unnecessarily-high IOPS.
Recurring quarterly audits find another 5-10% each time. Workloads grow; oversizing creeps back.
Automation tools (AWS Compute Optimizer, third-party FinOps platforms) automate the recommendation generation. Engineers approve and apply.
What to watch out for
Burst credit dynamics. gp2 uses I/O credits; bursts beyond baseline drain credits. Right-sizing storage may eliminate burst capacity needed during peaks.
Snapshot lineage. Snapshots reference the volume that produced them. Volume deletion after replacement should not orphan snapshot ancestry.
Filesystem expansion is one-way easy, contraction is hard. Plan for growth; right-size conservatively; don't aim for 100% utilisation.