Problem: Gitt 2 prosess I og J må du skrive et program som kan garantere gjensidig utelukkelse mellom de to uten ekstra maskinvarestøtte.
Avfall av CPU -klokkesykluser
I lekmannsbetegnelser da en tråd ventet på sin tur, endte den i en lang stund som testet tilstanden millioner av ganger per sekund og dermed gjorde unødvendig beregning. Det er en bedre måte å vente på, og det er kjent som 'avkastning' .
For å forstå hva det gjør trenger vi for å grave dypt inn i hvordan prosessplanleggeren fungerer i Linux. Ideen som er nevnt her er en forenklet versjon av planleggeren. Den faktiske implementeringen har mange komplikasjoner.
Tenk på følgende eksempel
Det er tre prosesser P1 P2 og P3. Prosess P3 er slik at den har en stundsløyfe som ligner på den i koden vår som ikke gjør så nyttig beregning, og den eksisterer fra løkken bare når P2 er ferdig med utførelsen. Planleggeren setter dem alle i en rund Robin -kø. Si nå at klokkehastigheten til prosessoren er 1000000/sek og den tildeler 100 klokker til hver prosess i hver iterasjon. Deretter blir først P1 kjørt for 100 klokker (0,0001 sekunder) og deretter P2 (0,0001 sekunder) etterfulgt av P3 (0,0001 sekunder) nå siden det ikke er flere prosesser denne syklusen gjentas til P2 slutter og deretter fulgt av P3s utførelse og til slutt avslutningen.
Dette er et fullstendig sløsing med 100 CPU -klokkesyklusene. For å unngå dette gir vi gjensidig opp CPU -tidskiven, dvs. utbytte som i det vesentlige ender denne gangskiven og planleggeren plukker opp neste prosess som skal kjøres. Nå tester vi tilstanden vår en gang så gir vi opp CPU -en. Tatt i betraktning testen vår tar 25 klokkesykluser sparer vi 75% av beregningen vår i en tidsskive. Å legge dette grafisk
Tatt i betraktning prosessorklokkehastigheten som 1MHz, er dette mye sparing!.
Ulike fordelinger gir forskjellig funksjon for å oppnå denne funksjonaliteten. Linux gir Sched_yield () .
void lock(int self) { flag[self] = 1; turn = 1-self; while (flag[1-self] == 1 && turn == 1-self) // Only change is the addition of // sched_yield() call sched_yield(); }
Minnegjerde.
Koden i tidligere opplæring kan ha fungert med de fleste systemer, men var ikke 100% riktig. Logikken var perfekt, men de fleste moderne CPU-er bruker ytelsesoptimaliseringer som kan resultere i utførelse utenfor orden. Denne omorganiseringen av minneoperasjoner (belastninger og butikker) går normalt upåaktet hen i en enkelt utførelsestråd, men kan forårsake uforutsigbar atferd i samtidige programmer.
Tenk på dette eksemplet
C
while (f == 0); // Memory fence required here print x;
I eksemplet ovenfor vurderer kompilatoren de to utsagnene som uavhengige av hverandre og prøver dermed å øke kodeeffektiviteten ved å bestille dem på nytt, noe som kan føre til problemer for samtidige programmer. For å unngå dette plasserer vi et minnegjerde for å gi hint til kompilatoren om det mulige forholdet mellom utsagnene på tvers av barrieren.
Så uttalelsesrekkefølgen
flagg [selv] = 1;
sving = 1-selv;
Mens (Turn Condition Check)
avkastning();
Må være nøyaktig den samme for at låsen skal fungere, ellers vil den havne i en dødvakttilstand.
For å sikre at disse kompilatorene gir en instruksjon som forhindrer bestilling av uttalelser over denne barrieren. I tilfelle av GCC ITS __sync_synchronize () .
Så den modifiserte koden blir
Full implementering i C:
// Filename: peterson_yieldlock_memoryfence.cpp // Use below command to compile: // g++ -pthread peterson_yieldlock_memoryfence.cpp -o peterson_yieldlock_memoryfence #include #include #include std::atomic<int> flag[2]; std::atomic<int> turn; const int MAX = 1e9; int ans = 0; void lock_init() { // Initialize lock by resetting the desire of // both the threads to acquire the locks. // And giving turn to one of them. flag[0] = flag[1] = 0; turn = 0; } // Executed before entering critical section void lock(int self) { // Set flag[self] = 1 saying you want // to acquire lock flag[self]=1; // But first give the other thread the // chance to acquire lock turn = 1-self; // Memory fence to prevent the reordering // of instructions beyond this barrier. std::atomic_thread_fence(std::memory_order_seq_cst); // Wait until the other thread loses the // desire to acquire lock or it is your // turn to get the lock. while (flag[1-self]==1 && turn==1-self) // Yield to avoid wastage of resources. std::this_thread::yield(); } // Executed after leaving critical section void unlock(int self) { // You do not desire to acquire lock in future. // This will allow the other thread to acquire // the lock. flag[self]=0; } // A Sample function run by two threads created // in main() void func(int s) { int i = 0; int self = s; std::cout << 'Thread Entered: ' << self << std::endl; lock(self); // Critical section (Only one thread // can enter here at a time) for (i=0; i<MAX; i++) ans++; unlock(self); } // Driver code int main() { // Initialize the lock lock_init(); // Create two threads (both run func) std::thread t1(func 0); std::thread t2(func 1); // Wait for the threads to end. t1.join(); t2.join(); std::cout << 'Actual Count: ' << ans << ' | Expected Count: ' << MAX*2 << std::endl; return 0; }
C // Filename: peterson_yieldlock_memoryfence.c // Use below command to compile: // gcc -pthread peterson_yieldlock_memoryfence.c -o peterson_yieldlock_memoryfence #include #include #include 'mythreads.h' int flag[2]; int turn; const int MAX = 1e9; int ans = 0; void lock_init() { // Initialize lock by resetting the desire of // both the threads to acquire the locks. // And giving turn to one of them. flag[0] = flag[1] = 0; turn = 0; } // Executed before entering critical section void lock(int self) { // Set flag[self] = 1 saying you want // to acquire lock flag[self]=1; // But first give the other thread the // chance to acquire lock turn = 1-self; // Memory fence to prevent the reordering // of instructions beyond this barrier. __sync_synchronize(); // Wait until the other thread loses the // desire to acquire lock or it is your // turn to get the lock. while (flag[1-self]==1 && turn==1-self) // Yield to avoid wastage of resources. sched_yield(); } // Executed after leaving critical section void unlock(int self) { // You do not desire to acquire lock in future. // This will allow the other thread to acquire // the lock. flag[self]=0; } // A Sample function run by two threads created // in main() void* func(void *s) { int i = 0; int self = (int *)s; printf('Thread Entered: %dn'self); lock(self); // Critical section (Only one thread // can enter here at a time) for (i=0; i<MAX; i++) ans++; unlock(self); } // Driver code int main() { pthread_t p1 p2; // Initialize the lock lock_init(); // Create two threads (both run func) Pthread_create(&p1 NULL func (void*)0); Pthread_create(&p2 NULL func (void*)1); // Wait for the threads to end. Pthread_join(p1 NULL); Pthread_join(p2 NULL); printf('Actual Count: %d | Expected Count:' ' %dn'ansMAX*2); return 0; }
Java import java.util.concurrent.atomic.AtomicInteger; public class PetersonYieldLockMemoryFence { static AtomicInteger[] flag = new AtomicInteger[2]; static AtomicInteger turn = new AtomicInteger(); static final int MAX = 1000000000; static int ans = 0; static void lockInit() { flag[0] = new AtomicInteger(); flag[1] = new AtomicInteger(); flag[0].set(0); flag[1].set(0); turn.set(0); } static void lock(int self) { flag[self].set(1); turn.set(1 - self); // Memory fence to prevent the reordering of instructions beyond this barrier. // In Java volatile variables provide this guarantee implicitly. // No direct equivalent to atomic_thread_fence is needed. while (flag[1 - self].get() == 1 && turn.get() == 1 - self) Thread.yield(); } static void unlock(int self) { flag[self].set(0); } static void func(int s) { int i = 0; int self = s; System.out.println('Thread Entered: ' + self); lock(self); // Critical section (Only one thread can enter here at a time) for (i = 0; i < MAX; i++) ans++; unlock(self); } public static void main(String[] args) { // Initialize the lock lockInit(); // Create two threads (both run func) Thread t1 = new Thread(() -> func(0)); Thread t2 = new Thread(() -> func(1)); // Start the threads t1.start(); t2.start(); try { // Wait for the threads to end. t1.join(); t2.join(); } catch (InterruptedException e) { e.printStackTrace(); } System.out.println('Actual Count: ' + ans + ' | Expected Count: ' + MAX * 2); } }
Python import threading flag = [0 0] turn = 0 MAX = 10**9 ans = 0 def lock_init(): # This function initializes the lock by resetting the flags and turn. global flag turn flag = [0 0] turn = 0 def lock(self): # This function is executed before entering the critical section. It sets the flag for the current thread and gives the turn to the other thread. global flag turn flag[self] = 1 turn = 1 - self while flag[1-self] == 1 and turn == 1-self: pass def unlock(self): # This function is executed after leaving the critical section. It resets the flag for the current thread. global flag flag[self] = 0 def func(s): # This function is executed by each thread. It locks the critical section increments the shared variable and then unlocks the critical section. global ans self = s print(f'Thread Entered: {self}') lock(self) for _ in range(MAX): ans += 1 unlock(self) def main(): # This is the main function where the threads are created and started. lock_init() t1 = threading.Thread(target=func args=(0)) t2 = threading.Thread(target=func args=(1)) t1.start() t2.start() t1.join() t2.join() print(f'Actual Count: {ans} | Expected Count: {MAX*2}') if __name__ == '__main__': main()
JavaScript class PetersonYieldLockMemoryFence { static flag = [0 0]; static turn = 0; static MAX = 1000000000; static ans = 0; // Function to acquire the lock static async lock(self) { PetersonYieldLockMemoryFence.flag[self] = 1; PetersonYieldLockMemoryFence.turn = 1 - self; // Asynchronous loop with a small delay to yield while (PetersonYieldLockMemoryFence.flag[1 - self] == 1 && PetersonYieldLockMemoryFence.turn == 1 - self) { await new Promise(resolve => setTimeout(resolve 0)); } } // Function to release the lock static unlock(self) { PetersonYieldLockMemoryFence.flag[self] = 0; } // Function representing the critical section static func(s) { let i = 0; let self = s; console.log('Thread Entered: ' + self); // Lock the critical section PetersonYieldLockMemoryFence.lock(self).then(() => { // Critical section (Only one thread can enter here at a time) for (i = 0; i < PetersonYieldLockMemoryFence.MAX; i++) { PetersonYieldLockMemoryFence.ans++; } // Release the lock PetersonYieldLockMemoryFence.unlock(self); }); } // Main function static main() { // Create two threads (both run func) const t1 = new Thread(() => PetersonYieldLockMemoryFence.func(0)); const t2 = new Thread(() => PetersonYieldLockMemoryFence.func(1)); // Start the threads t1.start(); t2.start(); // Wait for the threads to end. setTimeout(() => { console.log('Actual Count: ' + PetersonYieldLockMemoryFence.ans + ' | Expected Count: ' + PetersonYieldLockMemoryFence.MAX * 2); } 1000); // Delay for a while to ensure threads finish } } // Define a simple Thread class for simulation class Thread { constructor(func) { this.func = func; } start() { this.func(); } } // Run the main function PetersonYieldLockMemoryFence.main();
C++ // mythread.h (A wrapper header file with assert statements) #ifndef __MYTHREADS_h__ #define __MYTHREADS_h__ #include #include #include // Function to lock a pthread mutex void Pthread_mutex_lock(pthread_mutex_t *m) { int rc = pthread_mutex_lock(m); assert(rc == 0); // Assert that the mutex was locked successfully } // Function to unlock a pthread mutex void Pthread_mutex_unlock(pthread_mutex_t *m) { int rc = pthread_mutex_unlock(m); assert(rc == 0); // Assert that the mutex was unlocked successfully } // Function to create a pthread void Pthread_create(pthread_t *thread const pthread_attr_t *attr void *(*start_routine)(void*) void *arg) { int rc = pthread_create(thread attr start_routine arg); assert(rc == 0); // Assert that the thread was created successfully } // Function to join a pthread void Pthread_join(pthread_t thread void **value_ptr) { int rc = pthread_join(thread value_ptr); assert(rc == 0); // Assert that the thread was joined successfully } #endif // __MYTHREADS_h__
C // mythread.h (A wrapper header file with assert // statements) #ifndef __MYTHREADS_h__ #define __MYTHREADS_h__ #include #include #include void Pthread_mutex_lock(pthread_mutex_t *m) { int rc = pthread_mutex_lock(m); assert(rc == 0); } void Pthread_mutex_unlock(pthread_mutex_t *m) { int rc = pthread_mutex_unlock(m); assert(rc == 0); } void Pthread_create(pthread_t *thread const pthread_attr_t *attr void *(*start_routine)(void*) void *arg) { int rc = pthread_create(thread attr start_routine arg); assert(rc == 0); } void Pthread_join(pthread_t thread void **value_ptr) { int rc = pthread_join(thread value_ptr); assert(rc == 0); } #endif // __MYTHREADS_h__
Python import threading import ctypes # Function to lock a thread lock def Thread_lock(lock): lock.acquire() # Acquire the lock # No need for assert in Python acquire will raise an exception if it fails # Function to unlock a thread lock def Thread_unlock(lock): lock.release() # Release the lock # No need for assert in Python release will raise an exception if it fails # Function to create a thread def Thread_create(target args=()): thread = threading.Thread(target=target args=args) thread.start() # Start the thread # No need for assert in Python thread.start() will raise an exception if it fails # Function to join a thread def Thread_join(thread): thread.join() # Wait for the thread to finish # No need for assert in Python thread.join() will raise an exception if it fails
Produksjon:
Thread Entered: 1
Thread Entered: 0
Actual Count: 2000000000 | Expected Count: 2000000000