Question

A SaaS company runs a stateless, latency-sensitive REST API on a fleet of x86 On-Demand EC2 instances behind an Application Load Balancer. Traffic is predictable during business hours but spikes unpredictably during marketing campaigns. The application is written in Go and can be recompiled for any architecture. Leadership wants to reduce compute cost by at least 30% while maintaining consistent p99 latency during spikes and requiring no manual intervention when load changes. Which approach best meets these requirements?

Accepted Answer

Recompile the application for ARM64 and run it on Graviton-based instances in an EC2 Auto Scaling group with target tracking on ALB request count per target, using a mix of On-Demand and Spot capacity across multiple instance types.

Answer

Keep the x86 instances but purchase 3-year All Upfront Standard Reserved Instances sized for peak campaign traffic to guarantee capacity is always available.

Answer

Migrate the API entirely to Spot Instances in a single Auto Scaling group using one instance type, relying on Spot's lower price to achieve the cost target.

Answer

Move the API to a fixed-size cluster of larger x86 instances and enable ALB connection draining to smooth out traffic spikes.