tcmalloc ReleaseFreeMemory() not releasing the memory properly

Question

I'm using tcmalloc in one of my application in which the heap grow and shrink in very large amount, obviously I faced the issue where tcmalloc is not releasing the memory back to OS. Now I tried using the api to do that using MallocExtension::instance()->ReleaseFreeMemory();. It worked fine and released the memory. But when I keep my process running after some time (say 5 mins) the memory is still increasing to the initial level (sometimes more). The weird thing is application is idle.

Here is my code

#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include "google/malloc_extension.h"

int main(int argc, char* argv[])
{

    char** names;
    printf("\nBefore starting the execution. Press enter to start.... \n");
    getchar();
    if (argc < 3)
    {
        printf("Usage: ./a.out <numTimes> <allocsize>\n");
        exit(1);
    }
    int numTimes = atoi(argv[1]);
    int allocSize = atoi(argv[2]);
    names = (char**) malloc(numTimes * sizeof(char*));
    for (int i = 0; i < numTimes; i++)
    {
        names[i] = (char*)malloc(allocSize);
    }
    printf("\nDone with the execution. Press enter to free the memory.... \n");
    getchar();
    for (int i = 0; i < numTimes; i++)
    {
        free(names[i]);
    }
    free(names);
    printf("\nDone with the freeing. Press enter to release the memory.... \n");
    getchar();
    MallocExtension::instance()->ReleaseFreeMemory();
    printf("\nDone with the execution. Press enter to exit.... \n");
    getchar();
    return 0;
}



./a.out 10000 30000

after release

  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND            
18823 sarath    20   0  332m 4568 1268 S  0.0  0.2   0:00.05 a.out  

after sometimes(4-5 mins)

  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND            
18823 sarath    20   0  332m 129m 1268 S  0.0  6.5   0:00.05 a.out

Appreciate any help.

I gave up on tcmalloc and switched to tbbmalloc because tcmalloc in some cases doesn't free up memory. tbbmalloc offered similar performance and behaved better. — egur, Jan 01 '14 at 09:32
@Sarath You have an error here: `(char**)malloc(numTimes)` -> `malloc(numTimes*sizeof(char*))` otherwise memory not enought, you allocate only numTimes bytes but need to stored numTimes pointer (dword, i.e. 4 bytes each, as for x86 platform). Also your not freeing names itself after the free-loop. — mblw, May 01 '14 at 13:04
@Alexei Thanks for that. I was just concentrated on tcmalloc issue. Edited the question. — sarath, May 01 '14 at 13:32

score 3 · Answer 1 · answered Feb 27 '14 at 05:57

You can try including malloc.h and wrapping malloc_stats() around your call to MallocExtension::instance()->ReleaseFreeMemory(), like so....

malloc_stats();
MallocExtension::instance()->ReleaseFreeMemory();
malloc_stats();

You should then see something like this:

Before:

4997120 (    4.8 MiB) Bytes in page heap freelist
7434392 (    7.1 MiB) Actual memory used (physical + swap)
0 (    0.0 MiB) Bytes released to OS (aka unmapped)

After:

0 (    0.0 MiB) Bytes in page heap freelist
2437272 (    2.3 MiB) Actual memory used (physical + swap)
4997120 (    4.8 MiB) Bytes released to OS (aka unmapped)

If nothing else this will verify that the memory's actually being freed by from the page heap freelist and is now unmapped.

Here's my results Before MALLOC: + 316915712 ( 302.2 MiB) Bytes in page heap freelist MALLOC: = 330051736 ( 314.8 MiB) Actual memory used (physical + swap) MALLOC: + 0 ( 0.0 MiB) Bytes released to OS (aka unmapped) After MALLOC: + 0 ( 0.0 MiB) Bytes in page heap freelist MALLOC: = 13136024 ( 12.5 MiB) Actual memory used (physical + swap) MALLOC: + 316915712 ( 302.2 MiB) Bytes released to OS (aka unmapped) This is fine. But still after 4-5 mins the memory is growing up like the same. — sarath, Mar 04 '14 at 06:30

tcmalloc ReleaseFreeMemory() not releasing the memory properly

1 Answers1