What we learn running other people's cloud - cost, migrations, architecture and
the failures worth planning for.
Filter by tag
LinkedIn September 22, 2026
Flashback to January when EC2 Capacity Manager added way more info on Spot interruptions:
https://aws.amazon.com/about-aws/whats-new/2026/01/ec2-capacity-manager-spot-interruption-metrics/
Now that there’s been 9 months of metrics gathered for this feature, have you gone back and checked out the data? You might not have as high availability or cost savings as you thought.
Did you miss this week's blog post diving into secret ways Spot might NOT be saving money, even when Spot metrics claim great savings?
https://acloudabove.us/blogs/spot-secret-losses/
In high interruption AZs, check which instance types exist:
DescribeInstanceTypeOfferings API
aws ec2 describe-instance-type-offerings --region us-east-2 --location-type availability-zone --query 'InstanceTypeOfferings[].[InstanceType,Location]' --output text | sort | tr ' ' ','
If all your instance types are technically available in the high interruption AZ, it's possible that AZ just has much lower capacity, and you might need to swap out the instance type or AZ for another one.
Alternatively, if you’re using LowestPrice allocation strategy, try PriceCapcityOptimized. LowestPrice might launch an instance type with 1 left if its the lowest price, PCO balances capacity and price.
If launched through an ASG, EKS cluster, or many other services then the CreateFleet call in CloudTrail will give useful details. If you use a list of instance types (vs Attributes) the response section will detail every instance type Fleet tried to launch and couldn’t, with a specific reason.
Surprisingly, the answer could be No
Just one of the 5 losses here can reduce savings by HALF! (Hint: its the instance types)
Level: Advanced
Cost Optimization
Compute
EC2
Spot
Matched in the full text
LinkedIn September 16, 2026
Did you know IPv6 can often cost more than public IPv4 addresses in AWS?
An IPv4 header is generally 20B, while IPv6 is 40B. 20B difference means ~54 Million packets to get to 1GiB extra header data.
Billing differs by region and usage tier, but let’s pick $0.09/GiB for the first tier of traffic to the Internet from Ohio (us-east-2). Public IPv4 addresses cost ~5% that at $0.005/hour.
That means if your IP sends more than ~3 Million packets per hour out to the Internet, paying for IPv4 might actually be cheaper for you! For medium sized packet payloads(~1kB), that’s an instance pushing ~3GB of total payload data per hour, or ~7Mbps.
AWS pricing docs don’t directly state how headers are billed, but nothing indicates they aren’t. Tests from DuckBill Group and others have shown IPv4 headers *are* billed.
That “EC2 Other” line billing for IPv4 addresses might look annoying, but switching to IPv6 just hides the extra charge in your existing network line item.
Deleted an EBS volume and expected the billing to stop? Maybe not.
AWS introduced EBS Recycle Bin and rules could catch it leaving your full volume recoverable. That means that billing continues at the normal storage rate until retention expires or volume recovery.
Having a safety net is certainly useful, but left unchecked a wide retention rule could cost you big time.
Contact us today to learn about how we can help you tune and optimize cost for your EBS workload.
Are ALB rules costing you money a single config change could save?
If you have dozens of ALB listener rules, prioritize the most used ones. If “Rule Evaluations” is your highest LCU dimension, re-ordering the list reduces evaluations, since they stop at the first matching rule.
Does cost optimization always have to be a trade off with performance?
How do automated agents affect your spend?
And most importantly, what kind of ice cream is best paired with cost analysis?
Let's find out!
Level: Intermediate
Cost Optimization
Compute
Matched in the full text
LinkedIn August 24, 2026
This is a fun, if niche, new ASG feature that will help with API throttling where this workflow is used: https://aws.amazon.com/about-aws/whats-new/2026/08/amazon-ec2-auto-scaling-batch-termination
For workloads that need to selectively scale-in specific instances from a central control plane, you can now do that in bulk (100 instances per API call). Every ASG API shares a single throttle bucket, so this API wasn’t viable for larger workloads before.
For most use cases, ASG scaling policies work great, and Lifecycle Hooks can be used to gracefully shut down more stateful workloads. But for applications that need it, this is a welcome change! Do you have scaling questions you want to hear more about?
Book a free 30-minute consultation. Tell us where you are headed - migration, cost, architecture or resilience - and we will come back with a straight answer.
