5.5.26 · D1 · Coding › Embedded Systems & Real-Time Software › Fault tolerance — fail-safe vs fail-operational
Jo system kisi piece ke toot jaane par bhi kaam karta rahe, usse fault tolerant kehte hain. Ek toote hue piece se bachne ke sirf do seedhe tarike hain: safely ruk jaao (fail-safe) ya spare par chalta raho (fail-operational) — aur poora topic bas yeh measure karne ka math hai ki har ek kitne achhe se survive karta hai.
Is page par assume kiya gaya hai ki tumne kuch bhi nahi dekha . Parent note the parent topic ka har letter, har squiggle, har fraction yahan se ground up mein, usi order mein banaya gaya hai jis order mein tumhe chahiye. Kabhi aage mat kaodo — har symbol ko earn kiya jaata hai use karne se pehle.
Intuition Formula nahi, coin se shuru karo
Socho ki har ghante ek tiny invisible dice tumhare component ke liye roll hoti hai. Zyaadatar ghante woh kehti hai "theek hai". Bahut kam kabhi kehti hai "toot gaya". "Toot jaane" ka chance per hour ek fixed chhoti si number hoti hai. Hum us number ko ek naam dete hain taaki dobara "chance of breaking per hour" likhna na pade.
λ (lambda) — failure rate
λ ek Greek letter hai (bolo "LAM-da"). Iska matlab hai: average mein har ghante kitne failures hote hain .
Units: failures per hour, likha jaata hai h − 1 (padho "per hour").
Chhota λ = reliable part. Bada λ = fragile part.
Example: λ = 1 0 − 5 h − 1 matlab "average mein har 100 000 ghante mein ek failure".
Rate kyun, sirf "failing ka chance" kyun nahi? Kyunki failure time ke saath build up hoti hai. Jo part ek ghante mein aasaani se survive kare woh zaruri nahi ki hazaar ghante survive kare. Ek rate humein kisi bhi length of time ke baare mein poochhne deti hai. Neeche di gayi picture dekho.
Lal staircase time ke sath daayein jaate hue failures count karti hai. Iska average steepness exactly λ hai — symbol ka poora matlab yehi hai.
Parent note mein baar baar e − λ t aata hai. Aaiye iske har piece ko earn karte hain.
t — elapsed time
t simply kitni der se system chal raha hai , hours mein measure kiya, woh hai. t = 0 par sab kuch bilkul naya hai.
e
e ≈ 2.718 ek fixed number hai (jaise π ≈ 3.14 ). Yeh kisi bhi cheez ka natural partner hai jo time ke saath ek constant fraction se shrink hoti hai — jaise abhi-bhi-alive components ki population.
e − λ t "abhi bhi zinda fraction" kyun hai
1000 identical parts ko ek saath on kiya socho. Har ghante, survivors ka wahi chhota sa slice toot jaata hai. Fixed number of parts nahi — ek fixed fraction . Jab koi cheez bar bar ek fixed fraction se shrink hoti hai, toh survivor count e − λ t curve trace karta hai. Yehi WAHI akela curve hai jo aisa karta hai, isliye e aata hai.
Common mistake Minus sign ka confusion
Exponent negative hai: − λ t . Agar positive hota toh curve badhta , matlab parts khud hi theek ho jaate. Survival kabhi bhi sirf neeche jaati hai, isliye sign minus hona chahiye.
P ( kuch bhi )
P matlab probability hai: koi event kitna likely hai, 0 (kabhi nahi) se 1 (pakka) ke scale par. P = 0.5 matlab "50% chance". Brackets ke andar wala stuff event ka naam hai.
Toh P ( part still works at t ) = R ( t ) = e − λ t bas ek hi idea hai, do tarike se likha. Aur chance ki woh time t tak fail ho gaya hai, woh bacha hua sab hai:
Har failure equally buri nahi hoti. Ek signal jo accident se dark ho jaata hai (irritating par safe) alag hai uss ek se jo GREEN dikhata hai jab RED hona chahiye (deadly). Parent note λ ko do streams mein todta hai.
β (beta) — safe failure fraction
β (bolo "BAY-ta") 0 aur 1 ke beech ka plain fraction hai. Yeh jawaab deta hai: saari failures mein se kitna hissa harmless, detectable type ka hai?
β = 0.99 matlab 99% failures pakdi jaati hain aur safe bana di jaati hain; sirf 1% dangerous hain.
Ab total rate do chhhoti rates mein split hoti hai, jaise ek nadi do hissa ho jaati hai:
Split kyun karte hain? Kyunki fail-safe design sirf dangerous trickle λ d ki parwah karta hai. β ko 1 ke paas le jaane se woh trickle almost kuch nahi ho jaati — yahi fail-safe systems ki poori engineering game hai.
Worked example Dangerous stream ko squeeze karna
λ = 1 0 − 5 h − 1 aur β = 0.99 ke saath:
λ d = ( 1 − 0.99 ) × 1 0 − 5 = 0.01 × 1 0 − 5 = 1 0 − 7 h − 1
Dangerous rate 100 guna chhoti hai raw rate se. "Apni failures detect karo" yahi khareedta hai tumhare liye.
Parent note mein likha hai MTTDf = ∫ 0 ∞ e − λ d t d t . Do naye symbols: ∫ aur MTTDf.
Definition MTTF aur MTTDf — average lifetimes
MTTF = Mean Time To Failure = average number of hours ek part tootne se pehle chalta hai.
MTTDf = Mean Time To Dangerous Failure = specifically ek dangerous failure se pehle ke average hours.
"Mean" average ka hi doosra naam hai.
∫ 0 ∞ … d t
Yeh ek integral hai: ek machine jo infinity of tiny slices ko add up karti hai . ∫ 0 ∞ matlab "saare times ka sum 0 se forever tak". d t har time-slice ki tiny width mark karta hai.
average lifetime kyun deta hai
Pehle wali figure ki reliability curve R ( t ) socho. Uske neeche ka area total "survival ka amount" hai jo saare time mein accumulate hua. Woh area, magically, average lifetime ke barabar hota hai. Toh:
MTTF = ∫ 0 ∞ e − λ t d t = λ 1
Rate λ ke shrinking exponential ke neeche ka area exactly 1/ λ hota hai. Isliye average lifetime = rate ka ulta — fast rate (bada λ ) matlab chhoti life (chhota 1/ λ ), jo common sense se match karta hai.
Same machine ko dangerous stream par apply karo aur parent ka headline result milta hai:
MTTDf = ∫ 0 ∞ e − λ d t d t = λ d 1 = ( 1 − β ) λ 1
Fail-operational systems spares use karte hain. Kitne tarike se spares fail ho sakte hain yeh count karne ke liye ek aur notation chahiye.
( k n ) — "n choose k"
( k n ) (padho "n choose k ") count karta hai kitne tarike se n mein se k items pick kar sakte ho , order matter nahi karta.
( 1 3 ) = 3 : teen tarike ki kaun sa ek computer teen mein se fail hua.
( 2 3 ) = 3 : teen tarike ki kaun se do fail hue.
Kyun chahiye: teen identical units ke saath, "exactly ek fail hua" teen alag tarike se ho sakta hai, aur alag ways ki probabilities add up hoti hain. ( 1 3 ) wahi count hai.
Definition TMR — Triple Modular Redundancy
Teen identical units sab ek hi kaam karte hain. Ek voter majority answer leta hai. Agar ek unit galat kahe, baaki do use out-vote kar dete hain. System sirf tab harta hai jab 2 ya zyaada fail ho jaayein. (Aur padhne ke liye: Redundancy patterns (N+1, TMR, DMR) .)
Isliye parent TMR reliability ko ek sum ki tarah likhta hai: "teeno theek" plus "exactly ek kharab" — woh do situations jahan majority abhi bhi jeetti hai.
A ( T )
A = Availability = probability ki system kisi chosen moment par up aur working hai. A ( T ) mission-end time T par availability hai. "Four nines" matlab A > 0.9999 — das hazaar mein se ek se bhi kam hissa down.
Δ t m a x
Δ (Greek "delta") matlab "mein change" ya "itna amount". Δ t m a x maximum allowed time hai fault ke baad safe state reach karne ka. Isko miss karo toh "safe" bahut der se aata hai kuch kaam ka nahi. Dekho Watchdog timers and health monitoring .
ϵ
ϵ (Greek "epsilon") matlab ek tiny allowed slack — itni chhoti number ki hum treat karte hain "almost zero". ϵ = 1 0 − 9 matlab hum ek-in-a-billion imperfection accept karte hain, sabse uunche Safety Integrity Levels (SIL) ka bar.
S safe — safe state
Ek named condition jahan system fault hone par bhaag jaata hai: brakes clamped, signal RED, pump stopped. Subscript bas isse "safe" label karta hai. Yeh system ke behaviour mein ek jagah hai, number nahi.
reliability R equals e to minus lambda t
probability P from 0 to 1
MTTF equals 1 over lambda
integral adds tiny slices
split into lambda safe and lambda dangerous
Fail-operational availability A
Reveal karne se pehle jawab zor se bolo. Agar koi atka de, woh section dobara padho.
λ kya measure karta hai, aur kis units mein?Failure rate — average failures per hour, units h − 1 .
e − λ t kya represent karta hai?Reliability R ( t ) : probability ki ek part t ghante baad bhi kaam kar raha hai.
Exponent negative kyun hai? Survival sirf decrease hoti hai; positive exponent ka matlab hoga parts khud theek ho jaate.
R ( t ) ke terms mein P fail ( t ) kya hai?P fail ( t ) = 1 − e − λ t , kyunki alive + failed milke 1 hona chahiye.
β ka kya matlab hai, aur λ d kya hai?β woh fraction hai jo failures safe/detected hain; λ d = ( 1 − β ) λ dangerous failure rate hai.
Integral ∫ 0 ∞ e − λ t d t kya compute karta hai, aur kiske barabar hota hai? Reliability curve ke neeche ka area = mean time to failure = 1/ λ .
( 1 3 ) kितना equal hai aur TMR mein kyun aata hai?Yeh 3 ke barabar hai — kitne tarike se teen units mein se exactly ek failed ho sakta hai.
TMR system actually kab fail hota hai? Sirf tab jab 3 mein se 2 ya zyaada units fail ho jaayein, kyunki majority voter ek kharab unit ko override kar deta hai.
A ( T ) kya measure karta hai, aur "four nines" kya hai?Availability — probability ki system time T par operational hai; four nines matlab A > 0.9999 .
Δ t m a x aur ϵ kya hain?Δ t m a x safe state reach karne ki deadline hai; ϵ ek tiny accepted failure probability hai (jaise 1 0 − 9 ).