What does Fan-Out Tail Latency conclude, in numbers, about “When Independence Fails: Correlated Slowness, Bad Machines and Synchronised Background Work”?
Answer: Chapter 3 presents both directions of correlation: bad epochs lower the slow-page probability to 0.4033, fixed bad machines visited without replacement raise it to 0.6740, and synchronised background work presses it down to 0.0104.
Scope of the numbers: Service times, slow shares and costs are illustrative; all calculations are generated by the book's programs; the book is general business analysis, not individualised business advice. Marked boxes separate published evidence, runnable Python calculations (22 programs), classical sources and the author's hypotheses.
Data table
The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python onlyprinted values that match the answer: 0.4033
| marginal p = 0.01 per call, N = 100 calls per page | |||||
| w | pb | pg | rho | P(page slow) | slow/slow-page |
| 0.000 | 0.01 | 0.01000 | 0.0000 | 0.6340 | 1.58 |
| 0.005 | 0.50 | 0.00754 | 0.1219 | 0.5331 | 1.88 |
| 0.010 | 0.50 | 0.00505 | 0.2450 | 0.4033 | 2.48 |
| 0.010 | 0.90 | 0.00101 | 0.8082 | 0.1052 | 9.51 |
| 0.020 | 0.40 | 0.00204 | 0.3135 | 0.2011 | 4.97 |
| 0.050 | 0.10 | 0.00526 | 0.0431 | 0.4395 | 2.28 |
| 0.100 | 0.05 | 0.00556 | 0.0180 | 0.4838 | 2.07 |
| independence gives 0.6340; full correlation gives 0.0100 |
Show the full program
# Chapter 3. p31: correlated slowness that keeps the per-call rate fixed.
# Mixture model: a page falls in a "bad epoch" with probability w; then
# each of its N = 100 calls is slow with probability pb, otherwise with
# probability pg, where w pb + (1 - w) pg = 0.01 keeps every call's
# marginal slow rate at 0.01. rho is the intra-page correlation of the
# slow indicators. All parameter pairs are illustrative.
import math
P, N = 0.01, 100
def page_slow(p, n):
return -math.expm1(n * math.log1p(-p))
def mixture(w, pb):
pg = (P - w * pb) / (1.0 - w)
slow = w * page_slow(pb, N) + (1.0 - w) * page_slow(pg, N)
rho = w * (1.0 - w) * (pb - pg) ** 2 / (P * (1.0 - P))
return pg, slow, rho
if __name__ == "__main__":
print("marginal p = 0.01 per call, N = 100 calls per page")
print(" w pb pg rho P(page slow) slow/slow-page")
rows = ((0.0, 0.01), (0.005, 0.5), (0.01, 0.5), (0.01, 0.9),
(0.02, 0.4), (0.05, 0.1), (0.1, 0.05))
for w, pb in rows:
pg, slow, rho = mixture(w, pb)
per = N * P / slow
fmt = " {:5.3f} {:5.2f} {:7.5f} {:6.4f} {:6.4f}"
print(fmt.format(w, pb, pg, rho, slow) + " {:6.2f}".format(per))
print("independence gives {:.4f}; full correlation gives {:.4f}".format(
page_slow(P, N), P))Single source and citation
Fan, Y. (2026). The Multiplicative Law of Fan-Out Tail Latency and Hedged Requests: How Percentile Latency Amplifies When One Page Makes Many Calls (Version 1.0.0), Chapter 3: When Independence Fails: Correlated Slowness, Bad Machines and Synchronised Background Work. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-326 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026326,
author = {Fan, Yafu},
title = {The Multiplicative Law of Fan-Out Tail Latency and Hedged Requests: How Percentile Latency Amplifies When One Page Makes Many Calls},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 1 · The Multiplicative Law: How One Slow Call in a Hundred Becomes Sixty-Three Slow Pages in a Hundred
- Chapter 2 · The Page Is a Maximum: Order Statistics and the Last Few Leaves
- Chapter 4 · Hedged Requests: Sending a Second Copy at the 95th Percentile
- Chapter 5 · Queueing Tails: Utilisation Stretches the Tail, and Hedges Are Load Too
- Chapter 6 · Budget Allocation: Splitting a Page Percentile Across Every Call
- Chapter 7 · The Classics East and West: Half the Journey at Ninety, and the Law of the Minimum
- Chapter 8 · A Fan-Out Audit of a Paid Web Service, and a Plan for Refutation