SASADA Koichi <ko1 / atdot.net> wrote:
> (2014/05/06 4:47), Eric Wong wrote:
> > Right.  I think the problem was I ran "make check" on my server
> > and hit test_LSHIFT_neary_long_max in test/ruby/test_string.rb.
> > That test caused a lot of unused memory to be swapped out.
> 
> And solved after this fix? So the patch seems great!

Sorry for the confusion, the patch is good, but not a major fix.
I realized out a lot of RSS reduction was due to swap out
(from test_LSHIFT_neary_long_max running).

glibc malloc likes to keep memory and avoid releasing to the kernel,
so RSS mainly goes down from swapping (and not munmap or brk).

Maybe test_LSHIFT_neary_long_max is too slow/dangerous for
some systems.  I often force myself to use slower/older machines
to notice performance problems.