問題文
A service owner must publish a model behind an internal API. The business commitment is that 95 percent of requests return within a stated time, while the number of requests per second varies through the day. The owner is deciding what to optimize in the serving configuration. Which pair of properties is being balanced here?
選択肢
- Gradient accuracy and step count, because a serving endpoint continues to refine the weights as requests arrive.
- Rack power and cooling capacity, because those two determine how many requests the endpoint can accept.
- Checkpoint frequency and restart cost, because the environment resumes from the last saved state and the interval sets the work lost.
- Response time and throughput, because grouping requests together raises throughput but adds waiting time for the earliest request in the group.