关于「泊松到达下的溢出:桶深只平移漏桶队列的尾巴」,《私域速率整形》给出的定量结论是什么?

结论:看负载 0.60 的三行。纯漏桶时,每分钟结束仍有消息排队的概率是 0.048635,平均排队 0.0921 条,平均多等 0.01535 分钟;桶深 10 时排队概率降到 0.000008;桶深 20 时,六位小数里已看不到。负载 0.80 时,漏桶的排队概率是 0.281302,平均多等 0.10983 分钟;桶深 10 降到 0.004549,桶深 20 降到 0.000061。负载 0.90 时,漏桶的排队概率是 0.553500,平均排队 3.1012 条;桶深 10 时 0.074575;桶深 20 时 0.009397,平均多等 0.00558 分钟,不到半秒。

数据口径:示意计划每分钟平均 10.0 条,攒批猛发仍被拒收 420 条,接上桶深 20 的令牌桶后为零。 书中参数为示意,平台数字注明文档与查阅日期;全部计算由书中程序生成,程序不读写文件、不联网;本书是非个人化经营建议,不是法律或合规意见,任何线上发送节奏的改动须由主事者人工复核放行。

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: 0.000008, 0.000061, 0.004549, 0.00558, 0.009397, 0.01535

r = 10 per minute; Poisson load lam per minute
lamloadbP(queue at minute end)mean queuewait (min)
60.6000.0486350.09210.01535
60.60100.0000080.00000.00000
60.60200.0000000.00000.00000
80.8000.2813020.87860.10983
80.80100.0045490.01300.00162
80.80200.0000610.00020.00002
90.9000.5535003.10120.34458
90.90100.0745750.39860.04429
90.90200.0093970.05020.00558
查看完整验算程序
# Chapter 3. p31: Poisson arrivals through a token bucket, solved exactly.
# Model: event-driven messages (order notices, replies to customers who
# wrote first) arrive as Poisson(lam) per minute. The shaper adds
# r = 10 tokens a minute to a bucket of depth b. State s at the end of a
# minute = backlog - tokens, from -b upwards; next minute
# s' = max(s + A - r, -b). Put u = s + b: then u' = max(u + A - r, 0),
# the queue of a pure leaky bucket, whatever b is. So the backlog of a
# bucket of depth b is max(u - b, 0): one stationary law serves every
# depth. Law of u by iteration on 0 to 220. Mean wait beyond the arrival
# minute follows from Little's law: mean backlog / lam. Token-bucket
# arrival curves as in Le Boudec and Thiran, LNCS 2050.
import math

R, TOP = 10, 220


def poisson(lam, kmax=45):
    return [math.exp(-lam + k * math.log(lam) - math.lgamma(k + 1))
            for k in range(kmax)]


def leaky_law(lam, iters=6000):
    pa = poisson(lam)
    pi = [1.0] + [0.0] * TOP
    for _ in range(iters):
        bins = [[] for _ in range(TOP + 1)]
        for u, w in enumerate(pi):
            if w < 1e-300:
                continue
            for a, p in enumerate(pa):
                bins[min(max(u + a - R, 0), TOP)].append(w * p)
        new = [math.fsum(t) for t in bins]
        gap = max(abs(x - y) for x, y in zip(new, pi))
        pi = new
        if gap < 1e-15:
            break
    return pi


def at_depth(pi, b, lam):
    p_wait = math.fsum(pi[b + 1:])
    backlog = math.fsum((u - b) * w for u, w in enumerate(pi) if u > b)
    return p_wait, backlog, backlog / lam


if __name__ == "__main__":
    print("r = %d per minute; Poisson load lam per minute" % R)
    print("lam  load   b  P(queue at minute end)  mean queue  wait (min)")
    for lam in (6, 8, 9):
        pi = leaky_law(lam)
        for b in (0, 10, 20):
            pw, bl, wt = at_depth(pi, b, lam)
            print("%3d  %4.2f  %2d  %22.6f  %10.4f  %10.5f"
                  % (lam, lam / R, b, pw, bl, wt))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《私域通道防封号速率控制论、令牌桶限流与平台公示上限内的同意制触达吞吐母典》第 3 章「泊松到达下的溢出:桶深只平移漏桶队列的尾巴」(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-271 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026271,
  author    = {Fan, Yafu},
  title     = {Rate Control for Private Messaging Channels without Account Bans: Token-Bucket Shaping, Platform-Published Limits and Consent-Based Reach Throughput},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论