Exercise 2: Re-Measuring at N=1000 — Possible Solution
====================================================================

RESULTS
------------------------------
    REAL (file I/O + fsync), N=1000: 1305.64ms total -> 1305.64 microseconds/call
    FAKE (in-memory list), N=1000:    1.10ms total -> 1.10 microseconds/call
    Ratio: 1,185x

Compared to the chapter's own N=200 measurement (443x, 1387.04 vs.
3.13 microseconds/call), the per-call costs at N=1000 are similar in
absolute terms (real: ~1306 vs. ~1387 microseconds/call; fake: ~1.1 vs.
~3.1 microseconds/call), and the ratio - 1,185x here versus 443x in the
chapter - stays within the same broad order of magnitude (hundreds to
low thousands), even though the exact number moved.

WHY THIS WORKS AS AN ANSWER
------------------------------
The exact ratio isn't a fixed constant - it depends on real-world
factors like disk cache state and OS scheduling noise, which is
exactly why the chapter frames this as "the same order of magnitude,"
not an exact reproducible number. What stays consistent across both
runs is the underlying claim: real file I/O with a forced fsync is
consistently two-to-three orders of magnitude slower per call than an
in-memory operation, regardless of the specific sample size used to
measure it.