netdev - Re: [linux-next:master] [tcp] 8ee602c635: lmbench3.TCP.socket.bandwidth.10MB.MB/sec -21.8% regression

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <CANn89iJh7L_2b4YfMAcjrOF9-PSqTYA6XYNjjLmVFT8MBg60fg@mail.gmail.com>
Date: Tue, 30 Apr 2024 16:03:45 +0200
From: Eric Dumazet <edumazet@...gle.com>
To: kernel test robot <oliver.sang@...el.com>
Cc: oe-lkp@...ts.linux.dev, lkp@...el.com, 
	Linux Memory Management List <linux-mm@...ck.org>, Jakub Kicinski <kuba@...nel.org>, netdev@...r.kernel.org, 
	ying.huang@...el.com, feng.tang@...el.com, fengwei.yin@...el.com
Subject: Re: [linux-next:master] [tcp] 8ee602c635: lmbench3.TCP.socket.bandwidth.10MB.MB/sec
 -21.8% regression

On Tue, Apr 30, 2024 at 3:45 PM kernel test robot <oliver.sang@...el.com> wrote:
>
>
>
> Hello,
>
> kernel test robot noticed a -21.8% regression of lmbench3.TCP.socket.bandwidth.10MB.MB/sec on:
>
>
> commit: 8ee602c635206ed012f979370094015857c02359 ("tcp: try to send bigger TSO packets")
> https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git master
>
> [still regression on linux-next/master bb7a2467e6beef44a80a17d45ebf2931e7631083]
>
>
> testcase: lmbench3
> test machine: 48 threads 2 sockets Intel(R) Xeon(R) CPU E5-2697 v2 @ 2.70GHz (Ivy Bridge-EP) with 64G memory
> parameters:

I do not think I can spend time on this regression.

TCP performance is very sensitive to various factors, like packet
sizes, cache sizes,
sysctl settings (/proc/sys/net/ipv4/*), application design.

Making TSO packets bigger can increase p99 latencies, that is for sure.

>
>         test_memory_size: 50%
>         nr_threads: 100%
>         mode: development
>         test: TCP
>         cpufreq_governor: performance
>
>
>
>
> If you fix the issue in a separate patch/commit (i.e. not just a new version of
> the same patch/commit), kindly add following tags
> | Reported-by: kernel test robot <oliver.sang@...el.com>
> | Closes: https://lore.kernel.org/oe-lkp/202404302139.175c602f-oliver.sang@intel.com
>
>
> Details are as below:
> -------------------------------------------------------------------------------------------------->
>
>
> The kernel config and materials to reproduce are available at:
> https://download.01.org/0day-ci/archive/20240430/202404302139.175c602f-oliver.sang@intel.com
>
> =========================================================================================
> compiler/cpufreq_governor/kconfig/mode/nr_threads/rootfs/tbox_group/test/test_memory_size/testcase:
>   gcc-13/performance/x86_64-rhel-8.3/development/100%/debian-12-x86_64-20240206.cgz/lkp-ivb-2ep2/TCP/50%/lmbench3
>
> commit:
>   d5b38a71d3 ("tcp: call tcp_set_skb_tso_segs() from tcp_write_xmit()")
>   8ee602c635 ("tcp: try to send bigger TSO packets")
>
> d5b38a71d3334bc8 8ee602c635206ed012f97937009
> ---------------- ---------------------------
>          %stddev     %change         %stddev
>              \          |                \
>      50684           -21.8%      39620        lmbench3.TCP.socket.bandwidth.10MB.MB/sec
>      10.17 ą 41%     +83.6%      18.67 ą 15%  perf-c2c.HIT.remote
>       0.10 ą 71%      +0.1        0.22 ą 16%  perf-profile.self.cycles-pp.__tcp_push_pending_frames
>       0.66 ą 72%      +0.6        1.24 ą 14%  perf-profile.self.cycles-pp.tcp_write_xmit
>     965580 ą 83%    +150.9%    2422689 ą 51%  numa-meminfo.node1.FilePages
>    1617434 ą 52%     +91.0%    3089919 ą 40%  numa-meminfo.node1.MemUsed
>       3863 ą 14%     +24.5%       4810 ą 11%  numa-meminfo.node1.PageTables
>     612236 ą149%    +218.4%    1949210 ą 61%  numa-meminfo.node1.Unevictable
>     241348 ą 83%    +150.9%     605612 ą 51%  numa-vmstat.node1.nr_file_pages
>     964.36 ą 14%     +24.7%       1202 ą 10%  numa-vmstat.node1.nr_page_table_pages
>     153059 ą149%    +218.4%     487302 ą 61%  numa-vmstat.node1.nr_unevictable
>     153059 ą149%    +218.4%     487302 ą 61%  numa-vmstat.node1.nr_zone_unevictable
>   51738871 ą 15%     +27.3%   65841641        perf-stat.i.cache-misses
>       0.45 ą  7%      -7.5%       0.41        perf-stat.i.ipc
>       1.29 ą  2%     +17.0%       1.51        perf-stat.overall.MPKI
>       9.45            +1.9       11.39        perf-stat.overall.cache-miss-rate%
>       2.67            +3.2%       2.75        perf-stat.overall.cpi
>       2064 ą  2%     -11.8%       1820        perf-stat.overall.cycles-between-cache-misses
>       0.37            -3.1%       0.36        perf-stat.overall.ipc
>   51471548 ą 15%     +27.5%   65638078        perf-stat.ps.cache-misses
>
>
>
>
> Disclaimer:
> Results have been estimated based on internal Intel analysis and are provided
> for informational purposes only. Any difference in system hardware or software
> design or configuration may affect actual performance.
>
>
> --
> 0-DAY CI Kernel Test Service
> https://github.com/intel/lkp-tests/wiki
>