关于「分工表与通道修剪:网状、星形、小组与最长任务优先」,《团队规模的协调边界》给出的定量结论是什么?

结论:两个窗口时完工时间是三十八小时,加速比 1.97;三个窗口二十五小时,加速比 3.00,正好整除;四个窗口十九小时,加速比 3.95,每个窗口效率 0.99。到这里,分工几乎没有损失。五个窗口时最长任务优先给出十七小时,而理论下界是十五小时,加速比只有 4.41,效率降到 0.88,因为十二个长短不一的任务无法被五个人均匀分开。六个窗口又回到十三小时,效率 0.96;七个窗口十二小时,效率 0.89;八个窗口时,窗口的最大负荷降到十一小时,但协调者的十二小时审阅成了瓶颈,完工时间仍是十二小时,加速比停在 6.25,效率跌到 0.78。

数据口径:书中参数为示意;全部计算由书中程序生成;本书属于非个人化经营建议。 第四章以本出版局每波四卷并行为示意场景:容量峰值在 6.78 个窗口,四个窗口已拿到峰值的 89.6 %,协调者的审稿队列把上限压在四。

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: 0.78, 0.88, 0.89, 0.96, 0.99, 1.97

work 75 agent-hours, review 12 coordinator-hours
n LPT max lower bound makespan speed-up per agent
175 75 75 1.00 1.00
238 38 38 1.97 0.99
325 25 25 3.00 1.00
419 19 19 3.95 0.99
517 15 17 4.41 0.88
613 13 13 5.77 0.96
712 11 12 6.25 0.89
811 10 12 6.25 0.78
table for n4 (agent: volumes, hours):
agent 1: [9, 6, 4]19 h
agent 2: [8, 6, 5]19 h
agent 3: [8, 6, 5]19 h
agent 4: [7, 7, 4]18 h
查看完整验算程序
# Chapter 6. p62: a division-of-labour table by longest job first.
# Twelve volumes with illustrative writing times in agent-hours are
# assigned to n agents by the LPT rule (longest remaining job to the
# least loaded agent; ties to the lowest agent number). Each volume
# then needs 1 hour of the single coordinator's review, which cannot be
# split. Makespan >= max(LPT load, total review). Integers throughout.
import heapq
import math

JOBS = [9, 8, 8, 7, 7, 6, 6, 6, 5, 5, 4, 4]
REVIEW = 1


def lpt(jobs, n):
    heap = [(0, i, []) for i in range(n)]
    for j in sorted(jobs, reverse=True):
        load, i, got = heapq.heappop(heap)
        heapq.heappush(heap, (load + j, i, got + [j]))
    return sorted(heap, key=lambda t: t[1])


if __name__ == "__main__":
    work = int(math.fsum(JOBS))
    review = REVIEW * len(JOBS)
    print("work {} agent-hours, review {} coordinator-hours".format(
        work, review))
    print("   n  LPT max  lower bound  makespan  speed-up  per agent")
    for n in range(1, 9):
        loads = [t[0] for t in lpt(JOBS, n)]
        lb = max(-(-work // n), max(JOBS))
        span = max(max(loads), review)
        print("  {:2d}  {:7d}  {:11d}  {:8d}  {:8.2f}  {:9.2f}".format(
            n, max(loads), lb, span, work / span, work / span / n))
    print("table for n = 4 (agent: volumes, hours):")
    for load, i, got in lpt(JOBS, 4):
        print("  agent {}: {} -> {} h".format(i + 1, got, load))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《林格曼效应与布鲁克斯定律、团队与智能体军团规模的协调开销最优边界母典》第 6 章「分工表与通道修剪:网状、星形、小组与最长任务优先」(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-322 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026322,
  author    = {Fan, Yafu},
  title     = {The Ringelmann Effect and Brooks's Law: the Optimal Size of Teams and Agent Squads under Coordination Overhead},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论