关于「削峰填谷:短窗口重置下的延后、提前与积压池」,《周期配额的路由与倒计时》给出的定量结论是什么?

结论:第五章讲短窗口重置下的削峰填谷:只延后时产出 863.29 点,再允许例行任务提前升到 881.75 点,积压池把作废的配额变成产出。

数据口径:示意参数下,线性规划相对先到先用多出 0.1668 到 0.2217,相对按复杂度分档的固定规则只多出 0.41 % 到 5.05 %,大头来自分档本身。 书中参数为示意,不含任何厂商的价格或配额;全部计算由书中程序生成。

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: 863.29, 881.75

window123456
demand in F units43.086.0215.0258.0172.086.0
no shifting: points 787.50, idle quota units 327.4
H used100.0100.0100.0100.0100.0100.0
F used12.857.3150.0150.0145.257.3
defer only: points 863.29, idle quota units 229.9
H used100.0100.0100.0100.0100.0100.0
F used12.857.3150.0150.0150.0150.0
defer + pull: points 881.75, idle quota units 205.3
H used100.0100.0100.0100.0100.0100.0
F used117.8117.3150.0150.0129.630.0
peak-to-mean ratio of arriving F demand: 1.8000
查看完整验算程序
# Chapter 5. p51: peak shaving when the quota resets every window.
# Model: a day of 6 windows; the quota resets at the start of each
# window, H 100 and F 150 units per window (600 and 900 a day). Daily
# demand (300, 150, 60, 20) arrives with window shares 1, 2, 5, 6, 4, 2
# twentieths, so windows 3 and 4 are the peak. A task of a deferrable
# class may be served up to L windows after it arrives: L = 2 for
# routine and standard, 1 for complex, 0 for critical; nothing crosses
# the end of the day. A third case lets routine work that is known in
# advance (recurring jobs) also be pulled up to 2 windows earlier.
# One LP chooses window, shift and tier together.
# Same values, costs and success shares as p21. Units illustrative.
from fractions import Fraction as Fr

def simplex(c, A, b, rule="bland", cap=500):
    m, n = len(A), len(c)
    T = [[Fr(v) for v in A[i]] + [Fr(int(i == j)) for j in range(m)]
         + [Fr(b[i])] for i in range(m)]
    z = [-Fr(v) for v in c] + [Fr(0)] * (m + 1)
    basis = [n + i for i in range(m)]
    steps = 0
    while steps < cap:
        cols = [j for j in range(n + m) if z[j] < 0]
        if not cols:
            break
        if rule == "bland":
            e = cols[0]
        else:
            e = min(cols, key=lambda j: (z[j], j))
        rows = [i for i in range(m) if T[i][e] > 0]
        if not rows:
            return None
        r = min(rows, key=lambda i: (T[i][-1] / T[i][e], basis[i]))
        p = T[r][e]
        T[r] = [v / p for v in T[r]]
        for i in range(m):
            if i != r and T[i][e] != 0:
                f = T[i][e]
                T[i] = [a - f * w for a, w in zip(T[i], T[r])]
        f = z[e]
        z = [a - f * w for a, w in zip(z, T[r])]
        basis[r] = e
        steps += 1
    x = [Fr(0)] * (n + m)
    for i, j in enumerate(basis):
        x[j] = T[i][-1]
    return x[:n], z[-1], z[n:n + m], steps


VALUE = (1, 2, 5, 10)
COST = {"H": (4, 6, 10, 16), "F": (1, 2, 3, 4)}
SUCC = {"H": ("0.97", "0.95", "0.90", "0.85"),
        "F": ("0.95", "0.80", "0.45", "0.15")}
DAY = (300, 150, 60, 20)
SHARE = (1, 2, 5, 6, 4, 2)
W = 6
QW = {"H": 100, "F": 150}


def solve(lag):
    var = [(k, w, d, t) for k in range(4) for w in range(W)
           for d in range(lag[k][0], lag[k][1] + 1)
           if 0 <= w + d < W for t in "HF"]
    c = [VALUE[k] * Fr(SUCC[t][k]) for k, w, d, t in var]
    A, b = [], []
    for s in range(W):
        for t in "HF":
            A.append([COST[t][k] if (w + d == s and tt == t) else 0
                      for k, w, d, tt in var])
            b.append(QW[t])
    for k in range(4):
        for w in range(W):
            A.append([int(kk == k and ww == w) for kk, ww, d, t in var])
            b.append(Fr(DAY[k] * SHARE[w], 20))
    x, val, y, steps = simplex(c, A, b, rule="dantzig")
    load = {t: [Fr(0)] * W for t in "HF"}
    for (k, w, d, t), v in zip(var, x):
        load[t][w + d] += COST[t][k] * v
    return val, load


arr = [Fr(0)] * W
for w in range(W):
    for k in range(4):
        arr[w] += Fr(DAY[k] * SHARE[w], 20) * COST["F"][k]
print("window                   1      2      3      4      5      6")
print("demand in F units  " + "".join("%7.1f" % float(v) for v in arr))
CASES = (("no shifting", ((0, 0), (0, 0), (0, 0), (0, 0))),
         ("defer only", ((0, 2), (0, 2), (0, 1), (0, 0))),
         ("defer + pull", ((-2, 2), (0, 2), (0, 1), (0, 0))))
for name, lag in CASES:
    val, load = solve(lag)
    idle = Fr(0)
    for t in "HF":
        for v in load[t]:
            idle += QW[t] - v
    print("%s: points %.2f, idle quota units %.1f" % (
        name, float(val), float(idle)))
    for t in "HF":
        print("  %s used          " % t
              + "".join("%7.1f" % float(v) for v in load[t]))
tot = Fr(0)
for v in arr:
    tot += v
print("peak-to-mean ratio of arriving F demand: %.4f"
      % float(max(arr) * W / tot))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《周期配额削峰填谷与跨模型路由、重置倒计时线性规划与重推理轻量模型算力分配母典》第 5 章「削峰填谷:短窗口重置下的延后、提前与积压池」(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-270 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026270,
  author    = {Fan, Yafu},
  title     = {Peak-Shaving Periodic Quotas and Cross-Model Routing: Linear Programming against Reset Countdowns to Split Work between Heavy-Reasoning and Fast Models},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论