From: Eric Wong <normalperson@...>
Date: 2018-05-18T08:47:34+00:00
Subject: [ruby-core:87166] Re: [Ruby trunk Feature#14767] [PATCH] gc.c: use monotonic counters for objspace_malloc_increase

Eric Wong <normalperson@yhbt.net> wrote:
> OK.  Also, did we ever consider signed type (`ssize_t`) for malloc
> counters instead?  This may be more useful if we switch to
> thread-local storage for counters to avoid atomics.

Reverted for now because of regressions in:

            bm_array_sample_100k__6k
            bm_array_sample_100k___10k
            bm_array_sample_100k___50k

I think the original unsigned + underflow avoidance code
prevented us from accounting free() properly, and we were
triggering GC more as a result.

With this patch, frees are accounted properly so malloc_increase
grew too slowly...

Also, I think it would be beneficial to check malloc_increase
and do a lazy sweep step BEFORE calling malloc, since that should
improve cache locality in malloc (because they're usually LIFO).

Unsubscribe: <mailto:ruby-core-request@ruby-lang.org?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>