Visual walkthrough — Redundancy — cold standby, hot standby, active redundancy

3.6.32 · D2 · Physics › Spacecraft Structures & Systems Engineering › Redundancy — cold standby, hot standby, active redundancy

Hume probability se sirf ek idea chahiye aur kuch nahi. Chalo usse earn karte hain.

---

Step 1 — "Failure rate $\lambda$" ka asal matlab

KYA. Ek component random se fail hota hai. Hum describe karte hain ki woh kitni tezi se marne ki tendency rakhta hai — ek number se, jise $\lambda$ likhte hain (Greek letter "lambda"). Yeh failures per unit time ka count hai — maano, per hour.

IS number ko kyun use karein, koi aur kyun nahi. Hum keh sakte the "yeh exactly 5000 hours chalega." Lekin real hardware schedule pe nahi marta; woh randomly marta hai. Sabse honest cheez jo hum keh sakte hain woh hai ek rate: average mein, per hour kitne failures hote hain. Agar $\lambda =0.001$ per hour hai, toh ek ghante mein is part ke marne ki $0.001$ probability hai. Woh ek number baaki sab predict karne ke liye kaafi hai.

PICTURE. Neeche ki timeline dekho. Har chhoti tick ek "roll of the dice" hai per hour. Zyaadatar ghanton mein part survive karta hai; kabhi kabhi woh mar jaata hai. Bada $\lambda$ matlab death-events zyaada pass pass.

---

Step 2 — Rate se survival curve $e^{-\lambda t}$ tak

KYA. Hum chahte hain ki ek unit $t$ time ke baad abhi bhi zinda hone ki probability kya hai. Ise $R(t)$ kehte hain — reliability. Answer hai $R(t)=e^{-\lambda t}.$

Exponential kyun, seedhi line kyun nahi. Time ko chhoti chhoti $\Delta t$ width ki slices mein kato. Har slice mein part $(1-\lambda \Delta t)$ probability se survive karta hai — woh us slice mein death se bach jaata hai. Bahut saari slices survive karne ke liye un survival chances ko multiply karo. Bahut saare equal factors ko multiply karna exactly wahi hai jo ek exponential hota hai. Slices ko chhota karte jao aur product ban jaata hai $e^{-\lambda t}$. Yahi wajah hai ki exponential aata hai — yeh repeated survival hai, compounded.

Chalte hain symbols padh lete hain, jahan woh hain:

• $e$ — fixed constant $2.718\dots$, natural base jo "baar baar survival multiply karo" se nikalta hai.
• $-\lambda t$ — exponent. Yeh negative hai kyunki time badhne ke saath survive karna mushkil hota jaata hai. $\lambda$ scale karta hai ki kitni tezi se; $t$ hai kitna intezaar kiya.
• $t=0$ par: $R(0)=e^0=1$ (bilkul naya, pakka zinda). Jaise $t\to \infty$: $R\to 0$ (sab eventually marte hain).

PICTURE. Curve $1$ se shuru hota hai aur smoothly $0$ ki taraf decay karta hai. Steep $\lambda$ = tezi se decay.

Failure probability sirf bacha hua hai: $F(t)=1-e^{-\lambda t}$. Donon yaad rakho — do redundancy schemes inhe alag tarah se multiply aur add karti hain, aur wahi poori kahani hai.

---

Step 3 — Area trick: ek curve ko MTTF mein convert karna

KYA. Hume ek single headline number chahiye: Mean Time To Failure, average lifetime. Trick hai: $\text{MTTF}={\int }_0^{\infty }R(t)dt.$

Integral kyun, aur $R$ ka kyun. Symbol ${\int }_0^{\infty }\cdots dt$ ka matlab hai "$t=0$ se hamesha tak thin vertical strips of area add karo." Yahan ek khoobsoorat fact hai: survival curve ke neeche ka area average lifetime ke barabar hai. Intuition — har moment $t$ par, abhi zinda fraction $R(t)$ hai; ek unit jo $t$ par zinda hai woh apni khud ki lifespan mein ek aur instant $dt$ contribute karta hai. Poore time mein "alive-ness" ka sum sab ki lifetime total karta hai, average nikaal ke. Toh area = mean life. Humne integral isliye choose kiya kyunki yahi woh ek tool hai jo "kaun kab zinda hai" ko "average mein kitne time tak" mein convert karta hai.

PICTURE. $e^{-\lambda t}$ ke neeche shaded region MTTF hai. Ek single unit ke liye woh area clean $1/\lambda$ hai.

Check karo: ${\int }_0^{\infty }{e}^{-\lambda t}dt=\frac{1}{\lambda }$. Toh ek unit average mein $1/\lambda$ chalta hai. Agar $\lambda =0.001/\text{hr}$ hai, toh $1000$ hours. Yeh hamara baseline hai — woh number jise redundancy ko beat karna hai.

Recall MTTF,

$R(t)$ ke neeche area kyun hai? Kyunki $R(t)$ $t$ par abhi zinda fraction hai, aur us fraction ko poore time par sum karna (integral) sab lifespans ko ek average mein total karta hai. ::: Area under the survival curve = average lifetime.

---

Step 4 — HOT standby: do units saath aging kar rahe hain

KYA. Hot standby mein, dono units $t=0$ se powered hain. Dono age karte hain. System tab tak jinda hai jab tak kam se kam ek survive kare; woh tabhi marta hai jab dono mar jaayein.

Failure chances multiply kyun karein. "Dono dead" matlab unit-A dead AND unit-B dead. Independent parts ke liye, AND matlab multiply karo unki failure probabilities: $F_{\text{both}}(t)=(1-e^{-\lambda t})\times (1-e^{-\lambda t})=(1-e^{-\lambda t}{)}^2.$ Padhte hain: har factor $(1-e^{-\lambda t})$ hai "yeh ek unit $t$ tak pehle hi dead" Step 2 se; exponent $2$ hai "do independent units dono dead."

System reliability woh hai jo bacha hai: $R_{\text{hot}}(t)=1-(1-e^{-\lambda t}{)}^2=2e^{-\lambda t}-e^{-2\lambda t}.$

• $2e^{-\lambda t}$ — har unit ki survival ek baar count karo (unme se do hain).
• $-e^{-2\lambda t}$ — double-counted "dono zinda" overlap subtract karo taaki use do baar tally na karein.

PICTURE. Green two-unit curve single-unit curve ke upar hai — redundancy survival khareedti hai — lekin phir bhi $1$ se shuru hoti hai aur dono units apni clocks bilkul shuruat se jala rahe hain.

Ab area lo (Step 3 ki trick): ${\text{MTTF}}_{\text{hot}}={\int }_0^{\infty }(2e^{-\lambda t}-e^{-2\lambda t})dt=\frac{2}{\lambda }-\frac{1}{2\lambda }=\frac{3}{2\lambda }.$ $\frac{2}{\lambda }$ do survival terms se aata hai; $-\frac{1}{2\lambda }$ woh overlap hai jo tum subtract karte ho kyunki $e^{-2\lambda t}$ do guna tezi se decay karta hai (uska area aadha hai). Net: $\frac{3}{2\lambda }$ — dhaai lifetimes.

---

Step 5 — Sirf AADHA lifetime kyun, poora kyun nahi

KYA. Hot backup ne tumhe $\frac{3}{2\lambda }-\frac{1}{\lambda }=\frac{1}{2\lambda }$ diya — aadha lifetime. Poora nahi. Kyun?

KYU. Kyunki backup poore time powered on tha, parallel mein aging kar raha tha. Jis moment primary finally marta hai, hot backup apni khud ki zindagi ka ek hissa pehle hi spend kar chuka hota hai wahan hot aur running baithke. Tumhe ek partly-used spare milta hai, fresh nahi.

PICTURE. Do bars saath aging karte hue: jab primary bar khatam ho jaata hai, backup bar pehle hi partly consumed (shaded) hai. Bacha hua white piece bonus hai — sirf lagbhag aadha bar.

---

Step 6 — COLD standby: backup ki clock frozen hai

KYA. Cold standby mein backup OFF hai. Ek unpowered part barely age karta hai (Step 1 ke dice roll nahi ho rahe — no power, no wear). Uski lifetime clock sirf switchover par shuru hoti hai. Toh total system life primary ki life plus uske baad stack hua ek fresh full backup life hai.

Add kyun, overlap kyun nahi. Do lifetimes ab sequence mein hoti hain, kabhi same time mein nahi. Sequential independent lifetimes add hoti hain. System doosri failure par marta hai. Probability ki $t$ tak 2 se kam failures hue hain (yaani ab tak zero ya ek) woh Poisson survival ke pehle do terms hain: $R_{\text{cold}}(t)=\underset{\text{zero failures yet}}{\underbrace{e^{-\lambda t}}}+\underset{\text{exactly one failure, running on the spare}}{\underbrace{\lambda t{e}^{-\lambda t}}}=e^{-\lambda t}(1+\lambda t).$

• $e^{-\lambda t}$ — koi nahi mara; primary abhi chal raha hai.
• $\lambda t{e}^{-\lambda t}$ — exactly ek death hui aur hum ab fresh backup par ji rahe hain. $\lambda t$ count karta hai "kitna time spend hua," $e^{-\lambda t}$ survival envelope rakhta hai.

PICTURE. Cold curve (magenta) har jagah hot curve ke upar hai — frozen backup poore system ko zyaada time tak jelaata hai.

Phir area: ${\text{MTTF}}_{\text{cold}}={\int }_0^{\infty }{e}^{-\lambda t}(1+\lambda t)dt=\frac{1}{\lambda }+\frac{1}{\lambda }=\frac{2}{\lambda }.$ Pehla $\frac{1}{\lambda }$ primary ki poori life hai; doosra $\frac{1}{\lambda }$ (${\int }_0^{\infty }\lambda t{e}^{-\lambda t}dt$ se) backup ki poori life add hui. Do poori lifetimes.

---

Step 7 — Head-to-head comparison, aur jeet ki keemat

KYA. Inhe line up karo: ${\text{MTTF}}_{\text{cold}}=\frac{2}{\lambda }>\frac{3}{2\lambda }={\text{MTTF}}_{\text{hot}}>\frac{1}{\lambda }={\text{MTTF}}_{\text{single}}.$

Cold paper par kyun jeetta hai — lekin hamesha real life mein nahi. Cold reliability math jeetta hai kyunki uska spare fresh rehta hai. Cost time mein chupi hai: parent note ka switchover budget $t_{\text{detect}}+t_{\text{power-on}}+t_{\text{boot}}+t_{\text{sync}}$ cold ke liye minutes tak ja sakta hai, hot ke liye milliseconds. Mars entry ke dauraan, minutes fatal hain chahe MTTF bada ho. Toh choice hai reliability vs. downtime, exactly wahi tradeoff jo Reliability Block Diagrams aur Power Budget tumhe weigh karne par majboor karte hain.

PICTURE. Teeno survival curves stack ki huin, teeno shaded areas unke MTTF value se labelled hain.

---

Step 8 — Degenerate cases (reader ko cliff se mat girne do)

KYA & KYU & PICTURE — woh corners jo naive intuition ko todti hain:

• $t=0$ (bilkul naya). Har scheme $R=1$ deti hai. Dono curves summary figure par top-left mein saath shuru hoti hain — redundancy birth par koi head start nahi deta, sirf staying power deta hai. Summary figure par, saari curves $(0,1)$ par milti hain.

• $t\to \infty$ (lambi mission). Saari reliabilities $0$ tak decay karti hain — redundancy doom delay karta hai, kabhi khatam nahi karta. Yahi wajah hai ki deep-space missions phir bhi shorter hoti jaati hain jaise $\lambda t$ badhta hai.

• Imperfect switching ($p_{\text{switch}}<1$). Hamara cold-standby formula secretly assume karta tha ki switch hamesha kaam karta hai. Agar switch khud $1-p_{\text{switch}}$ probability se fail hota hai, cold ka advantage kam hota hai: $R_{\text{cold}}^{\text{real}}(t)=e^{-\lambda t}+p_{\text{switch}}\lambda t{e}^{-\lambda t}.$ Jab $p_{\text{switch}}=1$ toh yeh Step 6 ka result hai; jab $p_{\text{switch}}=0$ toh yeh single unit ($e^{-\lambda t}$) tak collapse ho jaata hai — backup useless hai kyunki tum usse kabhi reach nahi kar sakte. Yeh parent note ka "the switch is a single point of failure" hai, ab algebra mein visible. Hot standby, sirf $t_{\text{activate}}$ chahta hai, is se kam exposed hai — ek reason ki woh fast-switch cases mein FMEA scrutiny se bachta hai.

• Powered-but-idle wear (cold perfectly frozen nahi). Agar "off" backup actually reduced rate $\alpha \lambda$ se age karta hai jahan $0<\alpha <1$, cold ka MTTF $\frac{3}{2\lambda }$ ($\alpha =1$ par, effectively hot) aur $\frac{2}{\lambda }$ ($\alpha =0$ par, truly frozen) ke beech slide karta hai. Reality usi band mein rehti hai.

---

Ek-picture summary

Ek figure poora safar sameta hai: single-unit exponential (baseline area $1/\lambda$), uske upar hot curve (area $3/2\lambda$, backup saath aging kar raha hai), aur cold curve sabse upar (area $2/\lambda$, backup call hone tak frozen). Shaded areas ke beech gaps wahi reliability hai jo tumne kharidi — aur arrow woh keemat mark karta hai jo tum pay karte ho: switchover time.

Recall Feynman retelling — poora walkthrough simple alfazon mein

Ek part randomly marta hai; kitni tezi se — woh ek number hai, $\lambda$. ::: Survival har instant compound hoti hai, curve $e^{-\lambda t}$ deta hai. Average lifetime us curve ke neeche ka area hai. ::: Ek part ke liye woh area $1/\lambda$ hai. Hot standby dono parts ko shuruat se power karta hai, toh woh saath age karte hain; system sirf tabhi marta hai jab dono marte hain, matlab unke failure chances multiply karo. ::: Tumhe sirf aadha lifetime milta hai, $\frac{3}{2\lambda }$, kyunki spare hot baithte baithte aadha wear out ho gaya tha. Cold standby spare ko off rakhta hai, uski clock freeze karta hai; uski poori life primary ke baad stack hoti hai. ::: Tumhe poora lifetime milta hai, $\frac{2}{\lambda }$ — lekin tum minutes ka switchover time pay karte ho aur switch par trust karna padta hai.