关于「防卡顿:对数正态尾部、合并缓冲与含休息的专注预算」,《瓶颈定焦与价值工时比排程》给出的定量结论是什么?

结论:第一行是最乐观的情形:估计没有系统偏差,波动也小,离散参数 0.3。即便如此,不留缓冲的日子有 0.590 的概率超时;缓冲 0.15 时降到 0.231,缓冲 0.30 时降到 0.058,缓冲一半时只剩 0.006。第二行把波动放大到 0.6,不留缓冲的超时概率是 0.669,缓冲 0.30 时仍有 0.326,缓冲一半时 0.177。

数据口径:史密斯 1956 年证明,单人面对一张清单,使加权延迟代价最小的次序是按价值除以耗时排序,而不是按价值排序;七件示意任务上,按重要性排比穷举最优多付 0.311。 图形学的视锥剔除与细节层次说明,固定帧预算下不该每件都精做:八件示意任务全部精做只做完三件,总价值只有最优安排的 0.783。

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: 0.006, 0.058, 0.15, 0.177, 0.231, 0.3

reference optimism ratio 55.5 / 33.9 = 1.637
bsP(late) at buffer 00.150.300.50
1.0000.30.5900.2310.0580.006
1.0000.60.6690.4840.3260.177
1.2000.60.8610.7210.5650.375
1.6370.60.9860.9550.8950.771
mean / median of one task at s = 0.6: 1.197
查看完整验算程序
# Chapter 6. p61: log-normal task times and the chance a day overruns.
# A focus plan of six tasks estimated at 100, 70, 50, 40, 30, 10 minutes
# (300 in all). Actual minutes = estimate * b * exp(N(0, s^2)): b is a
# systematic optimism factor, s the spread. The reference ratio for b
# is Buehler, Griffin and Ross (1994): thesis students predicted 33.9
# days on average and took 55.5. A buffer of x adds x * 300 minutes.
# 50,000 simulated days per row; illustrative estimates.
import math
import random

EST = [100, 70, 50, 40, 30, 10]
PLAN = math.fsum(EST)
ratio = 55.5 / 33.9
print("reference optimism ratio 55.5 / 33.9 = %.3f" % ratio)


def overrun(b, s, buf, days=50000):
    rng = random.Random(37661)
    late = 0
    for _ in range(days):
        t = math.fsum(e * b * math.exp(rng.gauss(0.0, s)) for e in EST)
        if t > PLAN * (1.0 + buf):
            late += 1
    return late / days


print("b      s     P(late) at buffer 0    0.15    0.30    0.50")
for b, s in ((1.0, 0.3), (1.0, 0.6), (1.2, 0.6), (ratio, 0.6)):
    row = [overrun(b, s, x) for x in (0.0, 0.15, 0.30, 0.50)]
    print("%.3f  %.1f  %18.3f  %6.3f  %6.3f  %6.3f" % ((b, s) + tuple(row)))
mean_factor = math.exp(0.6 * 0.6 / 2.0)
print("mean / median of one task at s = 0.6: %.3f" % mean_factor)

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《阶段瓶颈动态定焦、价值工时比优先级排序与认知视锥渲染加速母典》第 6 章「防卡顿:对数正态尾部、合并缓冲与含休息的专注预算」(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-376 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026376,
  author    = {Fan, Yafu},
  title     = {Stage-Bottleneck Triage, Weighted Shortest-Processing-Time Scheduling and Level-of-Detail Execution Velocity},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论