A nightly feature-engineering job processes about 10 TB and takes six hours on a fleet of standard VMs. It tolerates restarts as long as completed partitions are kept. Which change lowers cost without changing the results?
Show answer and explanation
Answer: B
Spot VMs, formerly called preemptible VMs, cost substantially less than standard VMs and can be reclaimed at any time. Because the job is fault tolerant, saving completed partitions as checkpoints lets a replacement worker continue where the last one stopped, so reclaimed capacity costs little rework. Committed use discounts reward steady usage, but a job that runs a few hours each night would pay for idle committed capacity. Keeping the cluster running or shrinking it increases cost or time without addressing the price of the compute.
Why the other options are wrong
- A. A commitment discounts continuous usage, so for a six-hour nightly run most of the paid capacity would sit idle. It lowers the rate but raises total spend for this pattern.
- C. Running all month adds many hours of compute for no work, and the fixed setup cost is not reduced by leaving idle machines on. The bill grows substantially.
- D. Smaller machines extend the runtime to around 48 hours, which can miss the nightly window. Total compute hours may also rise, so the saving is unlikely to hold.