What does One-Site Thousand, Five-Site Matrix conclude, in numbers, about “The Price of a Shared Core: Correlated Failure, Blast Radius and the Chargeback Pool”?
Answer: A shared core does not change the average: under the illustrative parameters both architectures lose about RM 291 a month, and each site's availability is 0.99806 in both. What it changes is the tail: the one-per-cent level rises from RM 672.47 to RM 1,259.38, and in 0.696 of months the five sites stop together. Canary releases cut the loss from bad versions from RM 51.37 to RM 14.66 a month at almost no cost; bulkheads, timeouts and circuit breakers keep one failure in one place. The chargeback pool should be divided by risk: settled sites together, the new skin on its own.
Scope of the numbers: Order values, conversion rates, fees, incident rates and market parameters are illustrative, and all calculations are generated by the book's programs. Revenue figures are illustrative arithmetic, not a promise of income, and not investment, legal or tax advice.
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: 0.696, 0.99806, 1259.38, 672.47
| one shared core mean loss RM 291.27 a month, sd 275.21, p99 RM 1,259.38 | |
| months losing more than RM 1,000 | 0.0270 |
| five stacks mean loss RM 291.93 a month, sd 133.97, p99 RM | 672.47 |
| months losing more than RM 1,000 | 0.0000 |
| availability of one site in both designs | 0.99806 |
| months in which all five sites go down together | 0.696 with a shared core |
Show the full program
# Chapter 6. p61: five sites on one core against five separate stacks.
# Each site earns RM 1,000 a day spread evenly over the day. Core incidents
# arrive as Poisson with mean 1.2 a month and last an exponential 50 minutes
# on average; each skin has its own incidents, mean 0.8 a month, 30 minutes.
# With one shared core a core incident stops all five sites; with separate
# stacks each site has its own core with the same incident rate. Rates and
# durations are illustrative, not measurements of any provider.
import math
import random
import statistics
RNG = random.Random(357061)
PER_MIN = 1000.0 / 1440
MONTH_MIN, REPS = 30 * 1440, 20000
def poisson(lam):
limit, k, p = math.exp(-lam), 0, 1.0
while True:
p *= RNG.random()
if p <= limit:
return k
k += 1
def outages(rate, mean_min):
return [RNG.expovariate(1 / mean_min) for _ in range(poisson(rate))]
shared, separate, all_down = [], [], 0
for _ in range(REPS):
core = outages(1.2, 50.0)
skins = [outages(0.8, 30.0) for _ in range(5)]
skin_min = math.fsum(math.fsum(s) for s in skins)
shared.append(PER_MIN * math.fsum([5 * math.fsum(core), skin_min]))
own = math.fsum(math.fsum(outages(1.2, 50.0)) for _ in range(5))
separate.append(PER_MIN * math.fsum([own, skin_min]))
all_down += 1 if core else 0
for name, loss in (("one shared core", shared), ("five stacks", separate)):
loss.sort()
print("{:<16s} mean loss RM {:6.2f} a month, sd {:6.2f}, p99 RM {:8,.2f}".format(
name, statistics.fmean(loss), statistics.pstdev(loss),
loss[int(0.99 * REPS)]))
print(" months losing more than RM 1,000: %.4f" % (
len([x for x in loss if x > 1000]) / REPS))
avail = 1 - (1.2 * 50 + 0.8 * 30) / MONTH_MIN
print("availability of one site in both designs: %.5f" % avail)
print("months in which all five sites go down together: %.3f with a shared"
" core" % (all_down / REPS))Single source and citation
Fan, Y. (2026). The Thousand-Ringgit Single-Site Minimum Loop, Headless Core Multi-Skin Cloning and Five-Site Revenue Multiplication (Version 1.0.0), Chapter 6: The Price of a Shared Core: Correlated Failure, Blast Radius and the Chargeback Pool. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-357 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026357,
author = {Fan, Yafu},
title = {The Thousand-Ringgit Single-Site Minimum Loop, Headless Core Multi-Skin Cloning and Five-Site Revenue Multiplication},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 1 · Ask What a Thousand a Day Leaves: The Minimum Loop, Unit Contribution and Break-Even
- Chapter 2 · A Good Week Is Not a Stable Thousand: Daily Noise, Sequential Tests and the Clone Gate
- Chapter 3 · A Central Core Is Not Always Worth Building First: Headless Architecture, Information Hiding and Economies of Scope
- Chapter 4 · A New Skin Is Not a Copy of the Revenue: Per-Skin Accounts by Language, Country and Niche
- Chapter 5 · Five Sites That Do Not Raid Each Other: Orthogonal Partition of Keywords and Audiences, and the Multiplier
- Chapter 7 · East-West Classical Sources: Sunzi on Division and Variation, and Smith on the Extent of the Market
- Chapter 8 · System Architecture and Operations Spec: The 72-Hour Clone Pipeline, the Weekly Ledger and the Five-Site Matrix Cloning Calculator