[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20161130114339.78c6d4fa@redhat.com>
Date: Wed, 30 Nov 2016 11:43:39 +0100
From: Jesper Dangaard Brouer <brouer@...hat.com>
To: Rick Jones <rick.jones2@....com>
Cc: Eric Dumazet <eric.dumazet@...il.com>, netdev@...r.kernel.org,
brouer@...hat.com
Subject: Re: Netperf UDP issue with connected sockets
On Mon, 28 Nov 2016 10:33:49 -0800 Rick Jones <rick.jones2@....com> wrote:
> On 11/17/2016 12:16 AM, Jesper Dangaard Brouer wrote:
> >> time to try IP_MTU_DISCOVER ;)
> >
> > To Rick, maybe you can find a good solution or option with Eric's hint,
> > to send appropriate sized UDP packets with Don't Fragment (DF).
>
> Jesper -
>
> Top of trunk has a change adding an omni, test-specific -f option which
> will set IP_MTU_DISCOVER:IP_PMTUDISC_DO on the data socket. Is that
> sufficient to your needs?
The "-- -f" option makes the __ip_select_ident lookup go away. So,
confirming your new option works.
Notice the "fib_lookup" cost is still present, even when I use
option "-- -n -N" to create a connected socket. As Eric taught us,
this is because we should use syscalls "send" or "write" on a connected
socket.
My udp_flood tool[1] cycle through the different syscalls:
taskset -c 2 ~/git/network-testing/src/udp_flood 198.18.50.1 --count $((10**7)) --pmtu 2
ns/pkt pps cycles/pkt
send 473.08 2113816.28 1891
sendto 558.58 1790265.84 2233
sendmsg 587.24 1702873.80 2348
sendMmsg/32 547.57 1826265.90 2189
write 518.36 1929175.52 2072
Using "send" seems to be the fastest option.
Some notes on test: I've forced TX completions to happen on another CPU0
and pinned the udp_flood program (to CPU2) as I want to avoid the CPU
scheduler to move udp_flood around as this cause fluctuations in the
results (as it stress the memory allocations more).
My udp_flood --pmtu option is documented in the --help usage text (see below signature)
--
Best regards,
Jesper Dangaard Brouer
MSc.CS, Principal Kernel Engineer at Red Hat
LinkedIn: http://www.linkedin.com/in/brouer
$ uname -a
Linux canyon 4.9.0-rc6-page_pool07-baseline+ #185 SMP PREEMPT Wed Nov 30 10:07:51 CET 2016 x86_64
[1] udp_flood tool:
https://github.com/netoptimizer/network-testing/blob/master/src/udp_flood.c
Quick command used for verifying __ip_select_ident is removed:
# First run benchmark
sudo perf record -g -a ~/tools/netperf2-svn/src/netperf -H 198.18.50.1 \
-t UDP_STREAM -l 3 -- -m 1472 -f
# Second grep in perf output for functions
sudo perf report --no-children --call-graph none --stdio |\
egrep -e '__ip_select_ident|fib_table_lookup'
$ ./udp_flood --help
DOCUMENTATION:
This tool is a UDP flood that measures the outgoing packet rate.
Default cycles through tests with different send system calls.
What function-call to invoke can also be specified as a command
line option (see below).
Default transmit 1000000 packets per test, adjust via --count
Usage: ./udp_flood (options-see-below) IPADDR
Listing options:
--help short-option: -h
--ipv4 short-option: -4
--ipv6 short-option: -6
--sendmsg short-option: -u
--sendmmsg short-option: -U
--sendto short-option: -t
--write short-option: -T
--send short-option: -S
--batch short-option: -b
--count short-option: -c
--port short-option: -p
--payload short-option: -m
--pmtu short-option: -d
--verbose short-option: -v
Multiple tests can be selected:
default: all tests
-u -U -t -T -S: run any combination of sendmsg/sendmmsg/sendto/write/send
Option --pmtu <N> for Path MTU discover socket option IP_MTU_DISCOVER
This affects the DF(Don't-Fragment) bit setting.
Following values are selectable:
0 = IP_PMTUDISC_DONT
1 = IP_PMTUDISC_WANT
2 = IP_PMTUDISC_DO
3 = IP_PMTUDISC_PROBE
4 = IP_PMTUDISC_INTERFACE
5 = IP_PMTUDISC_OMIT
Documentation see under IP_MTU_DISCOVER in 'man 7 ip'
Powered by blists - more mailing lists