[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <20230726080916.17454-1-zhangpeng.00@bytedance.com>
Date: Wed, 26 Jul 2023 16:09:05 +0800
From: Peng Zhang <zhangpeng.00@...edance.com>
To: Liam.Howlett@...cle.com, corbet@....net, akpm@...ux-foundation.org,
willy@...radead.org, brauner@...nel.org, surenb@...gle.com,
michael.christie@...cle.com, peterz@...radead.org,
mathieu.desnoyers@...icios.com, npiggin@...il.com, avagin@...il.com
Cc: linux-mm@...ck.org, linux-doc@...r.kernel.org,
linux-kernel@...r.kernel.org, linux-fsdevel@...r.kernel.org,
Peng Zhang <zhangpeng.00@...edance.com>
Subject: [PATCH 00/11] Introduce mt_dup() to improve the performance of fork()
A few weeks ago, Liam and I discussed "Fork & Dup tree + Delete DONT_COPY" in
[1]. Thanks Liam for a lot of useful information there.
I didn't use the scheme of linking the leaf nodes, nor did I use the scheme
of building the tree in BFS order. I ended up using the scheme of building the
tree in DFS order. I implemented this algorithm and it seems to be very
efficient.
Use bench_forking() in lib/test_maple_tree.c to test in user space, and get the
performance numbers of duplicating maple tree as follows:
before: 13.52s
after: 2.60s
Their meaning is the time consumed by duplicating 80,000 maple trees with 134
entries. These numbers do not include the time consumed by mt_validate() and
mtree_destroy(). It can be seen that the time consumed has been reduced by
80.77%.
The performance improvement of fork() can be summarized as follows:
With 23 VMAs, performance improves by about 3%, with 223 VMAs, performance
improves by about 15%, and with 4023 VMAs, performance improves by about 30%.
See patch[11/11] for details.
In addition, I would like to assist Liam in maintaining the maple tree, which
requires Liam's consent. In the future I will make some contributions to the
development of maple tree.
The layout of these patches:
001 - 003: Introduce some internal functions to facilitate the
implementation of mt_dup().
004 - 005: Introduce __mt_dup() and mt_dup(), and their tests.
006: Introduce mas_replace_entry() to efficiently replace an entry.
007 - 009: Follow-up work on introducing these things.
010: Add myself as co-maintainer for maple tree.
011: Use __mt_dup() to duplicate maple tree in dup_mmap().
[1] https://lore.kernel.org/lkml/463899aa-6cbd-f08e-0aca-077b0e4e4475@bytedance.com/
Peng Zhang (11):
maple_tree: Introduce ma_nonleaf_data_end{_nocheck}()
maple_tree: Validate MAPLE_ENODE and ma_nonleaf_data_end()
maple_tree: Add some helper functions
maple_tree: Introduce interfaces __mt_dup() and mt_dup()
maple_tree: Add test for mt_dup()
maple_tree: Introduce mas_replace_entry() to directly replace an entry
maple_tree: Update the documentation of maple tree
maple_tree: Skip other tests when BENCH is enabled
maple_tree: Update check_forking() and bench_forking()
MAINTAINERS: Add co-maintainer for maple tree
fork: Use __mt_dup() to duplicate maple tree in dup_mmap()
Documentation/core-api/maple_tree.rst | 10 +
MAINTAINERS | 1 +
include/linux/maple_tree.h | 4 +
kernel/fork.c | 35 ++-
lib/maple_tree.c | 389 ++++++++++++++++++++++++--
lib/test_maple_tree.c | 67 ++---
mm/mmap.c | 14 +-
tools/testing/radix-tree/maple.c | 204 ++++++++++++++
8 files changed, 658 insertions(+), 66 deletions(-)
--
2.20.1
Powered by blists - more mailing lists