How cost functions as a design constraint

1. 01

   Every engineering choice has cost

   In traditional datacenters, hardware is a capital expense managed by procurement teams. In the cloud, developers create resources with API calls, making every engineering choice a financial choice. An inefficient query design or a wrong database index directly increases the monthly bill.

2. 02

   The three cost dimensions

   AWS pricing is built on three variables: compute time, storage volume, and network data transfer. While compute costs are easy to estimate from instance rates, network and storage costs are heavily dependent on access patterns and are frequently overlooked.

3. 03

   Data transfer is the hidden leak

   AWS charges for data leaving the cloud, data moving between Regions, and data moving between Availability Zones in the same Region. A multi-AZ system that constantly replicates database logs across zones can generate network transfer bills that exceed the compute costs.

1. 01

   Purchase models trade commitment for discounts

   On-Demand compute is the most expensive option. Reserved Instances and Savings Plans offer discounts in exchange for a commitment to a specific volume of compute usage for a one-year or three-year period. Spot Instances offer maximum discounts by selling spare capacity, but AWS can terminate them with a two-minute warning.

2. 02

   Idle resources charge by the hour

   Many AWS resources cost money even when they process zero requests. An Application Load Balancer, a NAT Gateway, and an unattached Elastic IP address all charge a flat hourly fee. If you launch these in multiple Availability Zones for development, you pay for idle infrastructure.

3. 03

   Storage classes affect access costs

   Amazon S3 offers multiple storage classes. Standard is cheap to write and read, but expensive to store. Glacier is cheap to store, but expensive to retrieve and requires hours to access. Placing frequently accessed data in Glacier can result in high retrieval charges.

4. 04

   Log retention defaults are infinite

   By default, Amazon CloudWatch Logs retains logs forever. A system that outputs gigabytes of debug logs daily will build up a massive archive over years. Without explicit log retention policies, your storage costs will grow linearly every month.

1. 01

   Match environment sizes to their usage

   Do not replicate your production environment's scale in development and staging. Development environments should use single-AZ deployments, smaller instance sizes, and auto-scaling groups configured to scale down to zero servers outside working hours.

2. 02

   Use VPC endpoints to bypass NAT

   If your private instances download gigabytes of data from S3, routing this traffic through a NAT Gateway incurs data processing charges. Creating a free S3 Gateway VPC Endpoint routes this traffic directly through the AWS network, eliminating NAT fees entirely.

3. 03

   Set up budget alarms on day one

   Never wait for the monthly invoice to discover a cost leak. Set up AWS Budgets with alarms that trigger when your actual or forecasted spend exceeds your budget by 10%. This gives you weeks of warning to terminate leaking resources.

Official references & further reading

These lessons simplify selected behaviors for learning. Verify current service limits, Region support and production requirements with the official references. Experiments describe their own assumptions.

• AWS Well-Architected: cost optimization