EBS Cost Optimization

EBS costs come from three sources: volume storage (per GB-month), IOPS (for io1/io2/io3), and snapshots (per GB-month). Most waste comes from over-sized volumes, old snapshots, and unattached volumes left behind after instance termination.

Volume Types and When to Use Them

Type	Use When	GB/Month	IOPS/GB	Max IOPS	Max Throughput
gp3	General purpose, most workloads	~$0.08	N/A (baseline 3,000 IOPS + 125MB/s included)	16,000	1,000 MB/s
gp2	Legacy, burst to 3,000 IOPS	~$0.10	Burst model	3,000	250 MB/s
io2	High-performance databases	~$0.125+$0.065 per IOPS	500:1 ratio	64,000	1,000 MB/s
io2 Block Express	Ultra-high performance	~$0.125+$0.065 per IOPS	1,000:1 ratio	256,000	4,000 MB/s
st1	Throughput-intensive (Hadoop, log processing)	~$0.045	N/A	500	500 MB/s
sc1	Cold storage (infrequently accessed)	~$0.015	N/A	250	250 MB/s

gp3 vs gp2: gp3 is newer and cheaper. It includes a baseline of 3,000 IOPS and 125MB/s regardless of volume size, and you can provision IOPS up to 16,000 independently of volume size. gp2 uses a burst model — small volumes get burst IOPS up to 3,000 but deplete a burst balance. gp3 is almost always the better choice for new workloads.

io2 vs gp3 for databases: io2 charges per provisioned IOPS. If your database needs 10,000 IOPS and you’re using io2 at $0.065/IOPS-month,tha{t}^{\prime }s$650/month just for IOPS on top of storage. gp3 at the same 10,000 IOPS is included in the price. io2 only makes sense when you need the higher durability (99.999%) or the multi-attach capability.

Unattached Volumes

The single biggest source of EBS waste is volumes left attached to stopped or terminated instances. EBS volumes persist independently of EC2 instances — when you stop or terminate an instance, the root volume is deleted (if DeleteOnTermination=true) but data volumes are detached and remain.

Finding unattached volumes:

aws ec2 describe-volumes --filters Name=status,Values=available --query 'Volumes[].[VolumeId,Size,VolumeType,CreateTime]'

Unattached volume detection strategies:

• AWS Config rule: ebs-volume-inuse-check detects volumes not attached to running instances
• AWS Trusted Advisor (Business plan): checks for unattached volumes
• Lambda function scheduled daily to report untagged unattached volumes
• AWS Instance Scheduler solution tags volumes with Duration and stops idle instances before volumes become orphaned

Snapshot Storage Costs

EBS snapshots are stored in S3. Snapshot pricing: ~$0.05/GB-month (standard, varies by region).

Sources of snapshot waste:

1. Old snapshots from deleted instances — the instance is gone but the snapshot remains
2. Multiple snapshots of the same volume — accumulated over time with no cleanup policy
3. Snapshots from test/dev environments — left behind after test runs
4. Encrypted snapshots with old KMS keys — can’t be deleted because the key is still referenced

Amazon Data Lifecycle Manager (DLM): Automate snapshot creation and deletion. Configure policies to:

• Create daily snapshots with a 7-day retention → auto-delete after 7 days
• Create weekly snapshots with 90-day retention → auto-delete after 90 days
• Tag-based policies for fine-grained control (only snapshot volumes with Backup=true tag)

DLM policies are free — you only pay for the snapshot storage.

Snapshot cleanup gotcha: DLM only manages snapshots it creates. Manually created snapshots need manual cleanup or a separate lifecycle policy.

Volume Sizing and Right-Sizing

Over-sized volumes: The most common issue. You provision 500GB because “we might need it” but actual usage is 50GB. You’re paying for 450GB of unused storage.

Under-sized volumes: Less common but causes performance issues. If IOPS requirements are high but volume is too small, you can’t provision enough IOPS (gp3 has a 50:1 IOPS-to-GB ratio for provisioned IOPS).

Right-sizing approach:

• CloudWatch VolumeConsumedReadWriteBytes metric shows actual I/O
• EBS I/O stats in the EC2 console shows if you’re consistently hitting volume limits
• Compute Optimizer recommends volume changes for instances with high EBS I/O

Multi-Attach Costs

io2 volumes support multi-attach (up to 16 instances). This is useful for clustered filesystems (GFS2, OCFS2) but adds complexity and cost. Multi-attach is almost never needed for application workloads — shared file storage should use EFS or FSx instead.

Encryption Costs

EBS encryption with AWS-managed keys (SSE-S3) is free — no additional charge. Customer-managed KMS keys incur charges only if you’re storing keys in a region with per-key charges ($1/month) or if you’re using CMK for volumes encrypted with a custom key (charged per volume-month for the key resource).

The only scenario where encryption adds cost: CMK with a dedicated HSM (CloudHSM), which has ongoing per-hour costs. Standard KMS CMK is effectively free for EBS encryption.

Monitoring EBS Costs

CloudWatch metrics for EBS:

• VolumeReadOps, VolumeWriteOps — I/O operations
• VolumeReadBytes, VolumeWriteBytes — throughput
• VolumeQueueLength — I/O waiting (high queue = volume is saturated)
• BurstBalance (gp2) — remaining burst IOPS credits

Use these to identify volumes that are:

• Consistently under-utilized (right-size down)
• Consistently over-utilized (upgrade to io2 or larger gp3)
• Showing low BurstBalance (gp2) indicating IOPS throttling

References

• Homepage: https://aws.amazon.com/ebs/pricing/
• Documentation: https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/AmazonEBS.html
• Pricing: https://aws.amazon.com/ebs/pricing/

Pricing Examples

Scenario 1: A database server using a500GB gp2 volume. Switching to gp3 with provisioned IOPS (500GB gp3 + 3000 IOPS): gp2 cost = 500GB × $0.10=$50/month. gp3 cost = 500GB × $0.08+3000IOPS\times$0.006 = $40+$18 = $58/month.Thegp3upgradecosts$8 more but provides consistent3000 IOPS vs burst gp2. For an IOPS-sensitive DB, worth it. For a lightly-used DB: gp3 at500GB with no extra IOPS = $40/month,saving$10/month.

Scenario 2: A development environment with 8 unattached EBS volumes (4 × 100GB gp2, 4 × 50GB gp2) sitting around from old test instances. Monthly cost: 8 × 100GB × $0.10=$80/month wasted. Deleting them saves $960/year. Set up a Lambda function to alert on unattached volumes older than 7 days and auto-delete after 30 days.

Nuggets & Gotchas

• gp2 burst IOPS are finite and deplete: gp2 volumes up to 1TB get 3 IOPS/GB burst capacity. A 100GB gp2 volume has 300 burst IOPS. If you sustain 400 IOPS, you burn through credits and drop to 300 baseline IOPS. gp3 eliminates this with consistent performance.
• EBS snapshots cost money even after you delete the volume: Snapshots are stored in S3 and charged per GB-month. A 500GB volume deleted but with 30 days of snapshots still costs $0.023/GB for those snapshots. Clean up old snapshots with lifecycle policies.
• io2 block express costs more but is cheaper at high IOPS: At > 64,000 IOPS per volume, io2 Block Express ($0.125/GB+$0.065/provisioned IOPS) becomes cheaper than io1 ($0.125/GB+$0.10/provisioned IOPS). For high-IOPS databases, do the math.
• EBS encryption is free (uses KMS CMK at no additional cost): There’s no cost for EBS encryption with the default AWS-managed CMK. Only custom CMKs cost $1/month. Don’t skip encryption for cost reasons.
• EBS volume performance is tied to the instance type: A nitro-based instance (e.g., c6i) can drive more IOPS from the same EBS volume than a non-nitro instance (e.g., c5). The volume type sets the ceiling, the instance type determines if you can reach it.