1 |
On Tue, 26 Feb 2013 08:33:44 -0500 |
2 |
Richard Yao <ryao@g.o> wrote: |
3 |
|
4 |
> The Blender project found some fairly remarkable discrepancies between |
5 |
> what their software actually used and what glibc's ptmalloc allocated: |
6 |
> |
7 |
> http://www.sintel.org/development/memory-jemalloc/ |
8 |
> |
9 |
> Results such as these led Blender and others (e.g. Chrome/Chromium, |
10 |
> Firefox, Thunderbird) to bundle private versions of jemalloc. |
11 |
|
12 |
Thre real disease is Windows support for all the mentioned apps. |
13 |
glibc is not that bad. |
14 |
|
15 |
I suspect |
16 |
export MALLOC_CHECK_=0 |
17 |
and |
18 |
CFLAGS=-fno-stack-protector |
19 |
would solve all those bookkeeping "issues" glibc dumps on you |
20 |
by default. |
21 |
|
22 |
malloc() (likely free()) performance dependency for an app |
23 |
means writing custom allocators. IMHO it's a per-app problem, |
24 |
not system-wide one. |
25 |
|
26 |
Do you have degradation example for your use cases? (a sample |
27 |
program for real workload) Fixing known deficiencies is always |
28 |
simpler than switching known-to-work base. |
29 |
|
30 |
> This bundling situation violates our policy against bundled libraries. The |
31 |
> maintainers could just patch their software to link to libjemalloc. |
32 |
> However, it might make more sense to evaluate jemalloc as a |
33 |
> distribution-wide replacement for glibc's ptmalloc. |
34 |
|
35 |
It would not solve bundling problem, would it? |
36 |
|
37 |
> Unless a significant issue is found in jemalloc itself, I do not see any |
38 |
> reason to continue using glibc's ptmalloc over jemalloc. |
39 |
|
40 |
How about "that core piece of libc works fine"? |
41 |
|
42 |
-- |
43 |
|
44 |
Sergei |