False sharing

1 minute read

Overview

False sharing is a performance-degrading usage pattern that can occur in systems with a distributed, coherent cache where the smallest resource block size is managed by caching mechanisms.

  • It refers to a condition where two processors attempt to write to different memory addresses that happen to map to the same cache line.
  • When processors repeatedly write to the same cache line, it triggers a substantial amount of communication between their caches due to the cache coherence protocol.

Example

#include <cstdio>
#include <chrono>
#include <pthread.h>

constexpr size_t
#if defined(__cpp_lib_hardware_interference_size)
  CACHE_LINE_SIZE = hardware_destructive_interference_size,
#else
  CACHE_LINE_SIZE = 64,
#endif
  MAX_THREADS = 8, MANY_ITERATIONS = 1000000000;

void* worker(void* arg) {
  volatile int* counter = (int*)arg;
  for (int i = 0; i < MANY_ITERATIONS; i++) (*counter)++;
  return NULL;
}
void test1(int num_threads) {
  auto begin = std::chrono::high_resolution_clock::now();

  pthread_t threads[MAX_THREADS];
  int counter[MAX_THREADS];

  for (int i = 0; i < num_threads; i++)
    pthread_create(&threads[i], NULL, &worker, &counter[i]);
  for (int i = 0; i < num_threads; i++)
    pthread_join(threads[i], NULL);

  auto end = std::chrono::high_resolution_clock::now();
  auto elapsed =
      std::chrono::duration_cast<std::chrono::nanoseconds>(end - begin);
  printf("Time measured: %.3f seconds.\n", elapsed.count() * 1e-9);
}

struct padded_t
{
  int counter;
  char padding[CACHE_LINE_SIZE - sizeof(int)];
};
void test2(int num_threads) {
  auto begin = std::chrono::high_resolution_clock::now();

  pthread_t threads[MAX_THREADS];
  padded_t counter[MAX_THREADS];

  for (int i = 0; i < num_threads; i++)
    pthread_create(&threads[i], NULL, &worker, &(counter[i].counter));
  for (int i = 0; i < num_threads; i++)
    pthread_join(threads[i], NULL);

  auto end = std::chrono::high_resolution_clock::now();
  auto elapsed =
      std::chrono::duration_cast<std::chrono::nanoseconds>(end - begin);
  printf("Time measured: %.3f seconds.\n", elapsed.count() * 1e-9);
}

int main()
{
    test1(8);
    test2(8);
}

Running the above code yields results similar to the following:

Time measured: 2.946 seconds.
Time measured: 2.533 seconds.

References

Leave a comment