💳 Ezezimali — ukusebenza kokukhokha Imodeli ephilayo

Isisekelo — imodeli engemuva kwezinombolo

Wonke ubudlelwano isilingisi esibusebenzisayo, nomthombo wabo. Izinto ezihlala zingashintshi ezimakwe njengokuthathelwa yizilinganiso zesibonelo.

Ukuthengiselana okusha kufika ngesilinganiso sesisekelo esiphindwe ukwenyuka kokuthengisa; inani ngomzuzu alihleliwe (Poisson, ukulinganisa okujwayelekile) nokuqhuma okuncane.

λ(t) = base × surge(t) (2-min lag); count/min ≈ N(60λ, √(60λ)) × (1 + N(0, 0.03))[2]Isimo esicatshangelwayo: inani lezikhala, izikhathi zensiza nezomkhiphi, ukwehlukana kwamaphuzu, amazinga okudlula e-step-up, ingxenye yomsebenzi osuke ungasebenzi, amanani nezindleko kuyizinombolo zesibonelo ze-switch esilinganiselayo, hhayi izibalo zanoma iyiphi inethiwekhi yangempela.

Iphuzu lokukhwabanisa elinamabili: amaphuzu eqiniso nawokukhwabanisa angamajika amabili ejwayelekile ahlukaniswe ngo-d′; umkhawulo ukhetha iphuzu ejikeni le-ROC.

genuine score ~ N(0,1), fraud ~ N(d′,1), d′ = 2.5; FPR(t) = 1 − Φ(t), TPR(t) = 1 − Φ(t − d′); AUC = Φ(d′/√2) ≈ 0.96[6][7]Isimo esicatshangelwayo: inani lezikhala, izikhathi zensiza nezomkhiphi, ukwehlukana kwamaphuzu, amazinga okudlula e-step-up, ingxenye yomsebenzi osuke ungasebenzi, amanani nezindleko kuyizinombolo zesibonelo ze-switch esilinganiselayo, hhayi izibalo zanoma iyiphi inethiwekhi yangempela.

Umkhawulo ongcono ngokwezindleko: yenqaba uma isilinganiso sokuthambekela sidlula isilinganiso sezindleko esilinganiswe izinga lesisekelo — ukukhwabanisa okuphindwe kayishumi kukwehlisa ngo-ln 10 / d′ ≈ 0.9σ.

decline when e^(d′x − d′²/2) ≥ (1 − π)·c_FP / (π·c_FN) ⇒ t* = [ln((1 − π)c_FP/(π c_FN)) + d′²/2] / d′[8][7]

I-step-up: amaphuzu ebhandeni elingaphansi komkhawulo anikezwa inselelo esikhundleni sokunqunywa.

scores in [t − b, t) are challenged: genuine pass 85 %, fraud pass 5 %; scores ≥ t declined[10]Isimo esicatshangelwayo: inani lezikhala, izikhathi zensiza nezomkhiphi, ukwehlukana kwamaphuzu, amazinga okudlula e-step-up, ingxenye yomsebenzi osuke ungasebenzi, amanani nezindleko kuyizinombolo zesibonelo ze-switch esilinganiselayo, hhayi izibalo zanoma iyiphi inethiwekhi yangempela.

Ukulibaza komkhiphi kuyisikhathi sokukhula ngokuphindaphindwa; isikhala siphethwe isikhathi sokulibaza noma sokuphela, noma yikuphi okuqala.

L ~ Exp(mean m); E[min(L, T)] = m(1 − e^(−T/m)), P(L > T) = e^(−T/m); S = 80 ms + forwarded × Σ share·E[min(L,T)][2][9]

Umthetho kaLittle: izikhala ezixakekile = izinga lokufika × isikhathi sokubamba.

busy slots = λ·S (Little) ⇒ utilization = λ·S / (32 × servers)[3]

I-Erlang C: ithuba lokuthi ukuthengiselana kulinde isikhala samahhala, nethuba lokuthi kulinde isikhathi eside kunokuphela kwesikhathi somugqa okungu-2 s.

P(W > 2 s) = C(a, N)·e^(−(N/S − λ)·2 s), a = λS; mean wait = C / (N/S − λ)[1][2]

Ngale komthamo i-switch iphinde isebenzise umsebenzi ezicelweni eziyozilahla kamuva, ngakho umphumela wokusebenza uyehla njengoba umthwalo ukhuphuka; ukwehlisa umthwalo kwenqaba okweqile ngokushibhile.

λ ≥ N/S: goodput = (N/S − ω·λ)/(1 − ω), ω = 0.3; load shedding admits ≤ 0.9·N/S and answers the rest at once[4][5]Isimo esicatshangelwayo: inani lezikhala, izikhathi zensiza nezomkhiphi, ukwehlukana kwamaphuzu, amazinga okudlula e-step-up, ingxenye yomsebenzi osuke ungasebenzi, amanani nezindleko kuyizinombolo zesibonelo ze-switch esilinganiselayo, hhayi izibalo zanoma iyiphi inethiwekhi yangempela.

Ukuzama futhi kwekhasimende: ukwehluleka ngakunye kwezobuchwepheshe kuyazanywa futhi kuze kube izikhathi ezintathu, ngokushesha noma nge-backoff yokukhula ngokuphindaphindwa okungahleliwe.

failed attempt → retry with p = 0.95, ≤ 3 retries; immediate: next minute; backoff: delay ~ Exp(mean 1, 2, 4 min)[4][5]Isimo esicatshangelwayo: inani lezikhala, izikhathi zensiza nezomkhiphi, ukwehlukana kwamaphuzu, amazinga okudlula e-step-up, ingxenye yomsebenzi osuke ungasebenzi, amanani nezindleko kuyizinombolo zesibonelo ze-switch esilinganiselayo, hhayi izibalo zanoma iyiphi inethiwekhi yangempela.

Isikhathi sokuphela somkhiphi, i-reversal ne-stand-in: amanani aku-lognormal, ngakho ingxenye engaphansi komkhawulo nevolumu evunyiwe kulandela kumsebenzi wokusabalalisa ojwayelekile.

issuer timeout → reversal; stand-in approves if amount ≤ limit: P = Φ((ln L − μ)/σ), volume = e^(μ+σ²/2)·Φ((ln L − μ − σ²)/σ)[9][11]

Izinga lokukhwabanisa kumaphoyinti esisekelo engxenyeni evunyiwe; u-13 bp usetshenziswa njengesikali sokubhekisa semiholo yamakhadi amancane akude.

fraud rate = approved fraud amount / approved amount × 10,000 bp (reference scale: 13 bp)[10]

Ezinye izinto ezingaguquki zokusebenza ezisetshenziswa yimodeli.

32 slots per server, 2–24 servers, +5 min boot, 6 u per server-hour · switch time 80 ms · queue timeout 2 s · issuer latency 250 ms (group B = 25 % of traffic, ±15 % per minute) · amounts lognormal, median 40 u, σ = 1, fraud ×1.5 · ω = 0.3 · shedding at 90 % · reversal = 80 ms of switch timeIsimo esicatshangelwayo: inani lezikhala, izikhathi zensiza nezomkhiphi, ukwehlukana kwamaphuzu, amazinga okudlula e-step-up, ingxenye yomsebenzi osuke ungasebenzi, amanani nezindleko kuyizinombolo zesibonelo ze-switch esilinganiselayo, hhayi izibalo zanoma iyiphi inethiwekhi yangempela.

Ukungahleliwe: i-generator ye-mulberry32 enembewu; ukusabalalisa okusetshenzisiwe — uniform, exponential (inverse CDF), normal (Box–Muller), Poisson (Knuth). Imbewu iyabonisa futhi ingabelwana.

Imithombo

1. Teletraffic Engineering Handbook — Erlang C formula; waiting-time distribution for M/M/n, FCFS — ITU-D Study Group 2 Question 16/2 (V. B. Iversen), 2005
2. M. Harchol-Balter — Performance Modeling and Design of Computer Systems: Queueing Theory in Action (M/M/k, capacity provisioning) — Cambridge University Press, 2013
3. J. D. C. Little — A Proof for the Queuing Formula: L = λW — Operations Research 9(3):383–387, 1961
4. Site Reliability Engineering — Ch. 22 Addressing Cascading Failures (retries, randomized exponential backoff, load shedding, work on requests past their deadline) — Beyer, Jones, Petoff, Murphy (eds.), O'Reilly, 2016
5. N. Bronson, A. Aghayev, A. Charapko, T. Zhu — Metastable Failures in Distributed Systems — HotOS ’21, ACM, 2021
6. T. Fawcett — An introduction to ROC analysis — Pattern Recognition Letters 27(8):861–874, 2006
7. R. J. Bolton, D. J. Hand — Statistical Fraud Detection: A Review — Statistical Science 17(3):235–255, 2002
8. C. Elkan — The Foundations of Cost-Sensitive Learning — IJCAI 2001, 2001
9. ISO 8583-1:2003 Financial transaction card originated messages — Interchange message specifications — Part 1 (authorization, reversal messages) — ISO/TC 68/SC 9, 2003
10. Commission Delegated Regulation (EU) 2018/389 — RTS on strong customer authentication, Annex: reference fraud rates (0.13 % for remote card payments up to EUR 100) — Official Journal of the European Union, 2018
11. N. L. Johnson, S. Kotz, N. Balakrishnan — Continuous Univariate Distributions, Vol. 1 (lognormal distribution, partial moments) — Wiley, 1994