Skip to main content

12 posts tagged with "finops"

View All Tags

We ran 89 cost checks on one AWS account: $32,000/year in findings, ranked

· 6 min read
Founder, Dragon Fractal · ex-AWS engineer

Short version: we pointed all 89 checks at one real AWS account, 1,358 resources in a single region. The scan returned 403 findings worth about $2,681/month, roughly $32,000/year, in potential savings. This is the raw surface area of waste a single account carries. It is not a promise you will bank all of it, and below I am going to be honest about why.

Graviton in 2026: the migration target moved to m8g, c8g, r8g

· 4 min read
Founder, Dragon Fractal · ex-AWS engineer

Short version: AWS Graviton is still about 20% cheaper than equivalent x86 and up to 40% better price/performance, and for most workloads it is a one-line instance-type swap. The thing that changed in 2026 is the target. Graviton4 (m8g, c8g, r8g) is now the current generation, up to 30% faster than Graviton3. If your last migration stopped at m6g or m7g, this is the update.

Case study: cutting $23,000/year from a production AWS bill, verified in Cost Explorer

· 6 min read
Founder, Dragon Fractal · ex-AWS engineer

Short version: we scanned a real production AWS account (~$10.7K/month, 1,084 resources in us-east-1), fixed six things, and cut $1,925.57/month. Every dollar was checked against AWS Cost Explorer billing data over three cycles. No production incidents. Here is what actually moved the bill.

Cut your AWS EBS bill ~20% by migrating gp2 to gp3

· 4 min read
Founder, Dragon Fractal · ex-AWS engineer

Short version: AWS gp3 EBS volumes cost ~20% less per GB than gp2 ($0.08 vs $0.10 per GB-month in us-east-1) and include 3,000 IOPS and 125 MB/s of baseline performance for free. You can convert a volume from gp2 to gp3 online, with no downtime and no snapshot: a single modify-volume call. For most volumes it is free money. Here is how to find them, migrate them, and the one caveat to check.

Idle load balancers: the ~$16/month each you forgot to delete

· 4 min read
Founder, Dragon Fractal · ex-AWS engineer

Short version: An Application or Network Load Balancer costs ~$0.0225/hour - about $16/month - just to exist, plus capacity units. Classic Load Balancers run ~$18/month. Load balancers outlive the services behind them: the app gets torn down, the ALB keeps billing. Here is how to find load balancers with no real traffic or no healthy targets, and remove them safely.

The $32/month surprise: finding and deleting idle NAT Gateways

· 4 min read
Founder, Dragon Fractal · ex-AWS engineer

Short version: Every NAT Gateway costs $0.045/hour, about $32/month, before a single byte of data, plus $0.045 per GB processed. They are easy to create per subnet and easy to forget. Here is how to find NAT Gateways that carry little or no traffic, delete them without breaking egress, and stop paying NAT data charges for traffic that could use free VPC endpoints instead.

Stop paying for unattached Elastic IPs and orphaned network interfaces

· 3 min read
Founder, Dragon Fractal · ex-AWS engineer

Short version: Since February 2024, AWS charges $0.005/hour for every public IPv4 address, about $3.60/month each, whether it is attached to anything or not. An unattached Elastic IP (EIP) is pure waste, and orphaned network interfaces (ENIs) often hold them. Individually small; at scale, and multiplied across accounts, it adds up. Here is how to find and release them safely.

Old EBS snapshots: the backup pile quietly inflating your bill

· 3 min read
Founder, Dragon Fractal · ex-AWS engineer

Short version: EBS snapshots cost $0.05/GB-month and nobody ever deletes them. Backups from volumes and AMIs that were removed years ago keep billing. For snapshots you must retain but rarely touch, the archive tier is ~75% cheaper ($0.0125/GB-month). Here is how to find the pile, delete what is dead, and archive what is not.

Idle Redshift clusters bill by the hour - even when nobody queries them

· 3 min read
Founder, Dragon Fractal · ex-AWS engineer

Short version: Amazon Redshift bills per node-hour whether the cluster runs a million queries or zero. A multi-node cluster left over from an old analytics project can run hundreds of dollars a month for nothing. Here is how to find idle clusters and stop the bleeding - pause, snapshot-and-delete, or move to Serverless.