diff --git a/src/zmalloc.c b/src/zmalloc.c
index 1cb01ee88c..e18fa8bac2 100644
--- a/src/zmalloc.c
+++ b/src/zmalloc.c
@@ -90,6 +90,7 @@ void zlibc_free(void *ptr) {
 
 #define thread_local _Thread_local
 
+#define PADDING_ELEMENT_NUM (CACHE_LINE_SIZE / sizeof(size_t) - 1)
 #define MAX_THREADS_NUM (IO_THREADS_MAX_NUM + 3 + 1)
 /* A thread-local storage which keep the current thread's index in the used_memory_thread array. */
 static thread_local int thread_index = -1;
@@ -101,10 +102,11 @@ static thread_local int thread_index = -1;
  * For the other architecture, lets fall back to the atomic operation to keep safe. */
 #if defined(__i386__) || defined(__x86_64__) || defined(__amd64__) || defined(__POWERPC__) || defined(__arm__) || \
     defined(__arm64__)
-static __attribute__((aligned(sizeof(size_t)))) size_t used_memory_thread[MAX_THREADS_NUM];
+static __attribute__((aligned(CACHE_LINE_SIZE))) size_t used_memory_thread_padded[MAX_THREADS_NUM + PADDING_ELEMENT_NUM];
 #else
-static _Atomic size_t used_memory_thread[MAX_THREADS_NUM];
+static __attribute__((aligned(CACHE_LINE_SIZE))) _Atomic size_t used_memory_thread_padded[MAX_THREADS_NUM + PADDING_ELEMENT_NUM];
 #endif
+static size_t *used_memory_thread = &used_memory_thread_padded[PADDING_ELEMENT_NUM];
 static atomic_int total_active_threads = 0;
 /* This is a simple protection. It's used only if some modules create a lot of threads. */
 static atomic_size_t used_memory_for_additional_threads = 0;