Read the following paper excerpts, and answer the questions below.
The total length of your response should not exceed ~500 words. Submit responses via gradescope.
Briefly describe the differences between open-loop, closed-loop, and partially-open-loop request generation.
The readings state that closed-loop designs can underestimate request latency in the tail. Why is this the case?
Section IIB of Treadmill states that:
one should extract the interested metrics (e.g., 99th-percentile latency) at each client individually, and aggregate them properly.
How exactly should the user “properly” aggregate these metrics? Construct an example set of results and show the process and results of your aggregation.
Describe an example of an application for which evaluating with a closed-loop request generator would be most appropriate. What would happen if one used an open-loop request generator to measure that application’s performance instead?