Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Mon, 29 Aug 2022 13:44:35 +0200
From:   Richard Gobert <richardbgobert@gmail.com>
To:     davem@davemloft.net, edumazet@google.com, kuba@kernel.org,
        pabeni@redhat.com, corbet@lwn.net, yoshfuji@linux-ipv6.org,
        dsahern@kernel.org, alex.aring@gmail.com,
        stefan@datenfreihafen.org, pablo@netfilter.org,
        kadlec@netfilter.org, fw@strlen.de, kafai@fb.com,
        netdev@vger.kernel.org, linux-doc@vger.kernel.org,
        linux-kernel@vger.kernel.org, linux-wpan@vger.kernel.org,
        netfilter-devel@vger.kernel.org, coreteam@netfilter.org
Subject: [PATCH 0/4] net-next: frags: add adaptive per-peer timeout under load
Message-ID: <20220829114427.GA2311@debian>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
User-Agent: Mutt/1.10.1 (2018-07-13)
Precedence: bulk

This patch series introduces an optimization of fragment queues under
load.

The goal is to improve upon the current approach of static timeouts
(frag_timeout, 30 seconds by default) by implementing Eric's
suggestion of reducing timeouts under load [1], with additional
considerations for peer-specific load.

The timeout reduction is done dynamically per peer, based on both
global and peer-specific load. low_thresh is reintroduced and now acts 
as a knob for adjusting per-peer memory limits.

A comparison of netperf results before and after applying the patch:

Before:
    [vm1 ~]# ./super_netperf.sh 10 -H 172.16.43.3 -l 60 -t UDP_STREAM
    103.23

After:
    [vm1 ~]# ./super_netperf.sh 10 -H 172.16.43.3 -l 60 -t UDP_STREAM
    576.17

And another benchmark of a more specific use case.
One high-bandwidth memory-hogging peer (vm1), and another "average"
client (vm2), attempting to communicate with the same server:

Before:
    [vm1 ~]# ./super_netperf.sh 10 -H 172.16.43.3 -l 60 -t UDP_STREAM
	42.57
	[vm2 ~]# ./super_netperf.sh 1 -H 172.16.43.3 -l 60 -t UDP_STREAM
	50.93

After:
    [vm1 ~]# ./super_netperf.sh 10 -H 172.16.43.3 -l 60 -t UDP_STREAM
	420.65
	[vm2 ~]# ./super_netperf.sh 1 -H 172.16.43.3 -l 60 -t UDP_STREAM
	624.79


These benchmarks were done using the following configuration:

[vm3 ~]# grep . /proc/sys/net/ipv4/ipfrag_*
/proc/sys/net/ipv4/ipfrag_high_thresh:104857600
/proc/sys/net/ipv4/ipfrag_low_thresh:78643200
/proc/sys/net/ipv4/ipfrag_max_dist:64
/proc/sys/net/ipv4/ipfrag_secret_interval:0
/proc/sys/net/ipv4/ipfrag_time:30

Regards,
Richard

[1] https://www.mail-archive.com/netdev@vger.kernel.org/msg242228.html

Richard Gobert (4):
  net-next: frags: move inetpeer from ip4 to inet
  net-next: ip6: fetch inetpeer in ip6frag_init
  net-next: frags: add inetpeer frag_mem tracking
  net-next: frags: dynamic timeout under load

 Documentation/networking/ip-sysctl.rst  |  3 +
 include/net/inet_frag.h                 | 13 ++---
 include/net/inetpeer.h                  |  1 +
 include/net/ipv6_frag.h                 |  3 +
 net/ieee802154/6lowpan/reassembly.c     |  2 +-
 net/ipv4/inet_fragment.c                | 77 ++++++++++++++++++++++---
 net/ipv4/inetpeer.c                     |  1 +
 net/ipv4/ip_fragment.c                  | 25 ++------
 net/ipv6/netfilter/nf_conntrack_reasm.c |  2 +-
 net/ipv6/reassembly.c                   |  2 +-
 10 files changed, 89 insertions(+), 40 deletions(-)

-- 
2.36.1