Exercise 2: 200ms Average and 8-Second p99 — Possible Solution ==================================================================== WHY BOTH CAN BE ACCURATE AT THE SAME TIME ------------------------------ Per this chapter, "average" is "a single blended figure - easily dominated by the bulk of fast, ordinary requests, hiding a smaller group of genuinely slow ones," while p99 shows "what the slowest 1% of requests actually experienced." If the large majority of requests are genuinely fast (say, well under 200ms) and a small minority (roughly 1%) are extremely slow (around 8 seconds), the average across all of them can still land near 200ms, since the fast majority mathematically dominates the blended figure - while the p99, looking specifically at the tail, accurately reports that the slowest 1% experienced something much worse. WHY THE AVERAGE ALONE MISSES THIS ------------------------------ The average treats every request as equally weighted toward one single number, which is exactly why it "hides" the slow minority - their extreme values get diluted across the much larger number of fast requests, the same mathematical effect Exercise 1 described for averaging over time, here applied across requests instead. WHY p99 MATTERS FOR DIAGNOSING COMPLAINTS ------------------------------ Per this chapter, the slowest 5% or 1% of requests are "exactly the group most likely to generate complaints." Users experiencing the 8-second p99 tail are the ones actually filing tickets, even while the 200ms average would suggest everything is fine - meaning the average is actively misleading about where the real, complained-about pain is coming from. WHY THIS WORKS AS AN ANSWER ------------------------------ It explains the specific mathematical reason both figures can be simultaneously accurate (a fast majority diluting a slow minority in the average), and connects the p99 figure directly to why it's more useful for diagnosing user complaints than the average is.