关于「翻倍的错觉:齐普夫请求下的命中率曲线」,《缓存命中率与车氏近似》给出的定量结论是什么?

结论:四个容量下,模拟命中率分别是 0.1380、0.2624、0.4496 与 0.7374,车氏近似给出的是 0.1373、0.2613、0.4483 与 0.7367,差距分别是 -0.0007、-0.0010、-0.0013 与 -0.0007,最大的绝对差距只有 0.0013。车氏近似没有任何从模拟里拟合出来的参数,它只用了请求概率和容量两样输入。对一个容量规划问题来说,这样的误差已经远小于请求概率本身的估计误差:第八章会看到,单是齐普夫指数估得偏一点,带来的误差就比这里大一个数量级。

数据口径:本书用标准库程序推翻这条直觉:在齐普夫请求下,最近最少使用(LRU)缓存的命中率随容量增长得很慢,示意参数下容量放大一百二十八倍,命中率只放大 6.80 倍;而车氏近似只凭请求概率与容量,就能事先算出 LRU 命中率,与逐条模拟的最大差距为 0.0013。 书中参数为示意;全部计算由书中程序生成;本书给出只读的容量建议,不访问任何服务器、不写任何配置;「排盘」仅作一种计算结果的名称,不做任何命理推断;属于非个人化经营建议。

English version

对照数据表

下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: -0.0007, -0.0010, -0.0013, 0.0013, 0.1373, 0.1380

LRU simulation vs Che, 250000 scored requests
CsimulatedCheChe - sim
1250.13800.1373-0.0007
5000.26240.2613-0.0010
20000.44960.4483-0.0013
80000.73740.7367-0.0007
largest absolute gap = 0.0013
查看完整验算程序
# Chapter 1. p13: a seeded LRU simulation against the Che prediction.
# Model: 300000 independent Zipf requests (a = 0.8, N = 20000 keys,
# illustrative) are replayed through a true LRU cache (OrderedDict);
# the first 50000 requests warm the cache and are not scored. The Che
# approximation is computed for the same p and C with no fitted
# parameter. Fixed seed 20261113, so every run prints the same table.
import math
import random
from collections import OrderedDict

N, A, L, WARM = 20000, 0.8, 300000, 50000


def zipf(n, a):
    w = [1.0 / (i ** a) for i in range(1, n + 1)]
    s = math.fsum(w)
    return [x / s for x in w]


def occupancy(p, t):
    return math.fsum(-math.expm1(-x * t) for x in p)


def che_t(p, c, tol=1e-9):
    lo, hi = 0.0, 1.0
    while occupancy(p, hi) < c:
        hi = 2.0 * hi
    while hi - lo > tol * hi:
        mid = 0.5 * (lo + hi)
        if occupancy(p, mid) < c:
            lo = mid
        else:
            hi = mid
    return 0.5 * (lo + hi)


def che_hit(p, t):
    return math.fsum(x * -math.expm1(-x * t) for x in p)


def lru_hit(trace, c, warm):
    cache = OrderedDict()
    hits = 0
    for t, k in enumerate(trace):
        if k in cache:
            cache.move_to_end(k)
            if t >= warm:
                hits += 1
        else:
            cache[k] = None
            if len(cache) > c:
                cache.popitem(last=False)
    return hits / (len(trace) - warm)


if __name__ == "__main__":
    p = zipf(N, A)
    rng = random.Random(20261113)
    trace = rng.choices(range(N), weights=p, k=L)
    print("LRU simulation vs Che, {} scored requests".format(L - WARM))
    print("      C   simulated    Che     Che - sim")
    worst = 0.0
    for c in (125, 500, 2000, 8000):
        s = lru_hit(trace, c, WARM)
        h = che_hit(p, che_t(p, c))
        worst = max(worst, abs(h - s))
        print("  {:5d}    {:.4f}    {:.4f}   {:+.4f}".format(c, s, h, h - s))
    print("largest absolute gap = {:.4f}".format(worst))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《缓存命中率与车氏近似、排盘结果缓存容量的最优边界母典》第 1 章「翻倍的错觉:齐普夫请求下的命中率曲线」(版本 1.0.0). 5D Library. DOI 待发布

DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-327 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026327,
  author    = {Fan, Yafu},
  title     = {Cache Hit Ratios and the Che Approximation: Optimal Capacity for Caching Computed Charts},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论