Question

A media analytics company runs a nightly batch job that transcodes and analyzes user-uploaded video. The workload is highly parallelizable, fault-tolerant (individual tasks can be retried), and must complete within a 6-hour window each night. The rest of the day the compute fleet sits idle. Job volume varies significantly from night to night. The finance team wants to minimize compute cost while ensuring the batch reliably finishes within its window. Which compute strategy should the solutions architect recommend?

Accepted Answer

Use AWS Batch with a compute environment that combines Spot Instances via a diversified allocation strategy across multiple instance types, and configure a fallback to On-Demand capacity if Spot capacity becomes unavailable.

Answer

Provision a fixed fleet of Reserved Instances sized for peak nightly volume so capacity is always guaranteed and reserved pricing lowers cost.

Answer

Run the batch on a single large On-Demand EC2 instance with a Savings Plan applied, scaling vertically as job volume grows.

Answer

Deploy the workload on an Auto Scaling group of On-Demand Instances behind a queue, using scheduled scaling to launch capacity every night at a fixed size.