EC2 Purchasing Option for a 3-Hour Weekly Workload That Cannot Be Interrupted?

Why the Answer Is Correct

On-Demand Instances are the AWS-recommended purchasing option for exactly this profile: "short-term, irregular workloads that cannot be interrupted." Three hours per week is roughly 12–13 instance-hours per month, which is neither steady-state nor a capacity-reservation scenario, so paying per second with no upfront fee and no term commitment matches both the cost and the no-interruption requirements. Because there is no commitment, the company cannot end up paying for unused compute if the research schedule changes or the lab pauses. The answer also satisfies the constraint that the workload "cannot be interrupted": On-Demand capacity is not reclaimed by AWS the way Spot capacity is.

Why the Other Options Are Wrong

A Compute Savings Plan does offer flexible discounts, but it requires a 1- or 3-year commitment to a specific hourly spend, and the discount only pays off when usage is consistent and predictable; a lab burning 3 hours a week would have to commit at a negligible level, and any commitment above actual usage is money spent on nothing. Convertible Reserved Instances are even less suitable, since they lock a 1- or 3-year term tied to instance attributes for steady-state workloads, and the flexibility to exchange them does not change the commitment problem. Spot Instances are eliminated outright by the requirement itself: AWS can reclaim Spot capacity with a two-minute interruption notice, which directly violates "cannot be interrupted," regardless of how cheap the capacity is.

Community Comment Notes

The community leaned strongly toward B, and Yak_Yeti summarized the reasoning by describing On-Demand as ideal for workloads that run for short durations and cannot be interrupted. Naantje reinforced this by quoting the AWS EC2 User Guide line that On-Demand is recommended for "short-term, irregular workloads that cannot be interrupted," which is precisely the scenario described here. On the other side, kimbi argued for a Compute Savings Plan because the lab's usage is predictable and repeated weekly, and Marley insisted that "MOST cost-effective" must mean the discounted plan; both arguments treat the word "cost-effective" as synonymous with "discounted" and ignore the tiny, non-steady footprint. Floridian noted the roughly 60/40 split in votes and asked whether the answer key settles it — the tie-breaker is the AWS guidance, not the vote count.