[Bug libc/11261] malloc uses excessive memory for multi-threaded applications

["rich at testardi dot com" <sourceware-bugzilla at sourceware dot org>]

------- Additional Comments From rich at testardi dot com  2010-02-10 13:41 -------
Hi Ulrich,

Agreed 100% no one size fits all...

Unfortunately, the neither of the "tuning" settings for MALLOC_ARENA_MAX nor 
MALLOC_ARENA_TEST seem to work.  Neither do mallopt() M_ARENA_MAX nor 
M_ARENA_TEST. :-(

Part of the problem seems to stem from the fact that the global "narenas" is 
only incremented if MALLOC_PER_THREAD/use_per_thread is true...

#ifdef PER_THREAD
  if (__builtin_expect (use_per_thread, 0)) {
    ++narenas;

    (void)mutex_unlock(&list_lock);
  }
#endif

So the tests of those other variables in reused_arena() never limit anything.  
And setting MALLOC_PER_THREAD makes our problem much worse.

static mstate
reused_arena (void)
{
  if (narenas <= mp_.arena_test)
    return NULL;

  ...

  if (narenas < narenas_limit)
    return NULL;

I also tried all combinations I could imagine of MALLOC_PER_THREAD and the 
other variables, to no avail.  I also did the same with mallopt(), verifying 
at the assembly level that we got all the right values into mp_. :-(

Specifically, I tried things like:

export MALLOC_PER_THREAD=1
export MALLOC_ARENA_MAX=1
export MALLOC_ARENA_TEST=1

and:

    rv = mallopt(-7, 1);
    printf("%d\n", rv);
    rv = mallopt(-8, 1);
    printf("%d\n", rv);

Anyway, thank you.  You've already pointed me in all of the right directions.  
If I did something completely brain-dead, above, feel free to tell me and save 
me another few days of work! :-)

-- Rich

-- 


http://sourceware.org/bugzilla/show_bug.cgi?id=11261

------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.