Serverless Computing Questions and Answers

Answer: C

Duration is usually measured in milliseconds and weighted by the memory configured. An idle function costs nothing at all, which is what makes the model attractive for spiky or infrequent workloads. The memory setting therefore appears in the price as well as in the performance.

Answer: D

One hundred invocations at two tenths of a second each comes to twenty seconds of execution. Set against a continuously running instance billed for eighty six thousand four hundred seconds, the economics of the model become obvious.

Answer: A

Correct answer: Both A and R are true, and R is the correct explanation of A

Both are true and the reason gives the crossover. Serverless bills per invocation and per gigabyte-second, which is very cheap when the function runs occasionally and idles the rest of the time. Under continuous load those charges accumulate past the flat cost of a reserved instance of equivalent capacity. The model suits spiky and unpredictable work, not steady throughput.

Answer: B

The first two are correct and together describe where the crossover lies. The third is the overstatement, and comparing the two at the actual expected utilisation is the only way to settle it. The answer differs by an order of magnitude between a spiky and a steady load.

Answer: C

Correct answer: the number of invocations executing at the same instant

Concurrency is a measure of simultaneity, not of volume. A function called a million times a month, each taking 50 milliseconds, may never exceed a concurrency of two; one called a thousand times an hour, each taking a minute, sustains a concurrency of about seventeen. Since limits are set on concurrency, it is the product of rate and duration that matters.

Answer: D

Correct answer: Both A and R are true, and R is the correct explanation of A

Both are true and the reason gives the mechanism. Throttling triggers when the number of simultaneous executions reaches the limit, and a burst of arrivals can reach that limit even if each invocation finishes in milliseconds. The remedies are to raise the concurrency limit, to smooth the arrivals through a queue, or to reduce the duration - since concurrency is rate multiplied by duration.

Answer: A

More memory makes each invocation faster but does nothing about how many may run at once. The other three each either raise the ceiling or spread the arrivals out over time. Absorbing a burst is a question of admission, not of how quickly each call finishes.

Answer: B

Correct answer: False

Function platforms impose a maximum execution time, commonly fifteen minutes, so a two-hour job cannot complete in a single invocation. Such work belongs in a container service, a batch service or a workflow that breaks it into steps - and the limit is not an arbitrary restriction but part of what makes the pricing and scheduling model work.

Answer: C

Correct answer: an image thumbnail generated when a file is uploaded

Thumbnail generation is the archetypal fit: event-driven, short, stateless, independently parallel and idle most of the time. The other three are all poor fits - consistently low latency conflicts with cold starts, a large resident model must be reloaded into each new environment, and a multi-hour job exceeds the execution limit.

Answer: D

Correct answer: A-2, B-4, C-1, D-3

Four independent limits. The timeout caps how long one invocation may run. The memory setting determines the resources each invocation gets. The concurrency limit caps how many may run at once. And the payload size caps how large a single event may be. Each is hit in a different way, which is why diagnosing a failure starts with identifying which of the four was reached.