Commit 26006d7e authored by Kirill Smelkov's avatar Kirill Smelkov

go/neo/t/neotest: Network information & benchmarks

Add to neotest bench-net command that performs latency measurments at
ping and TCP levels. Example output:

	x/src/lab.nexedi.com/kirr/neo/go/neo/t$ ./neotest bench-net neotest@rio:9
	node:
	cluster:        deco-rio

	*** link latency:

	# deco ⇄ rio (ping 16B)
	PING rio (192.168.0.8) 16(44) bytes of data.

	--- rio ping statistics ---
	25705 packets transmitted, 25705 received, 0% packet loss, time 2999ms
	rtt min/avg/max/mdev = 0.080/0.097/0.220/0.011 ms, ipg/ewma 0.116/0.095 ms
	Benchmarkpingrtt-/16B-min 1 0.080 ms/op
	Benchmarkpingrtt-/16B-avg 1 0.097 ms/op
	# POLL·3 C1·476 C1E·60917 C3·53 C6·132 C7s·0 C8·203 C9·0 C10·141

	...

	*** TCP latency:

	# deco ⇄ rio (lat_tcp.c 1B  -> lat_tcp.c -s)
	Benchmarktcprtt(c_c)-/1B 1 116.1743 µs/op       # TCP latency using rio: 116.1743 microseconds  # POLL·6 C1·892 C1E·65748 C3·80 C6·165 C7s·0 C8·339 C9·0 C10·444
	Benchmarktcprtt(c_c)-/1B 1 117.2896 µs/op       # TCP latency using rio: 117.2896 microseconds  # POLL·4 C1·1063 C1E·67647 C3·64 C6·77 C7s·0 C8·144 C9·0 C10·209
	Benchmarktcprtt(c_c)-/1B 1 117.5331 µs/op       # TCP latency using rio: 117.5331 microseconds  # POLL·1 C1·954 C1E·76866 C3·96 C6·88 C7s·0 C8·206 C9·0 C10·246
	Benchmarktcprtt(c_c)-/1B 1 117.6509 µs/op       # TCP latency using rio: 117.6509 microseconds  # POLL·4 C1·731 C1E·84210 C3·103 C6·93 C7s·0 C8·180 C9·0 C10·187
	Benchmarktcprtt(c_c)-/1B 1 116.8125 µs/op       # TCP latency using rio: 116.8125 microseconds  # POLL·9 C1·550 C1E·79544 C3·110 C6·213 C7s·0 C8·508 C9·0 C10·475

	...

And its summary via benchstat:

	name                 time/op
	pingrtt-/16B-min     80.0µs ± 0%
	pingrtt-/16B-avg     97.0µs ± 0%
	-pingrtt/16B-min     79.0µs ± 0%
	-pingrtt/16B-avg      112µs ± 0%
	pingrtt-/1452B-min    241µs ± 0%
	pingrtt-/1452B-avg    303µs ± 0%
	-pingrtt/1452B-min    266µs ± 0%
	-pingrtt/1452B-avg    303µs ± 0%
	tcprtt(c_c)-/1B       117µs ± 1%
	tcprtt(c_go)-/1B      122µs ± 2%
	-tcprtt(c_c)/1B       117µs ± 1%
	-tcprtt(c_go)/1B      121µs ± 5%
	tcprtt(c_c)-/1400B    392µs ± 4%
	tcprtt(c_go)-/1400B   363µs ±18%
	-tcprtt(c_c)/1400B    412µs ±21%
	-tcprtt(c_go)/1400B   391µs ±38%
	tcprtt(c_c)-/1500B    271µs ±18%
	tcprtt(c_go)-/1500B   290µs ±21%
	-tcprtt(c_c)/1500B    282µs ±16%
	-tcprtt(c_go)/1500B   334µs ±24%
	tcprtt(c_c)-/4096B    711µs ± 5%
	tcprtt(c_go)-/4096B   737µs ± 5%
	-tcprtt(c_c)/4096B    740µs ± 2%
	-tcprtt(c_go)/4096B   711µs ± 7%

Latencies here are not good because for this run on rio interrupt mitigation
was not tuned (see below). By the way, analyzing ping RTT latencies on our
shuttle machines (similar to rio) resulted in the following kernel patch

	https://git.kernel.org/linus/509708310c (released with Linux 4.15)

to fix/being able to adjust interrupt mitigation on Realtek NICs.

While at networking topic, teach info/info-local to show related
information about node's NICs. Example lines output for deco:

	nic/eth0: Intel Corporation Ethernet Connection I219-LM rev 21
	nic/eth0/features: rx tx sg tso !ufo gso gro !lro rxvlan txvlan !ntuple rxhash ...
	nic/eth0/coalesce: rxc: 3μs/0f/0μs-irq/0f-irq,  txc: 0μs/0f/0μs-irq/0f-irq
	nic/eth0/status:   up, speed=1000, mtu=1500, txqlen=1000, gro_flush_timeout=0.000µs
	nic/wlan0: Intel Corporation Wireless 8260 rev 3a
	nic/wlan0/features: !rx !tx sg !tso !ufo gso gro !lro !rxvlan !txvlan !ntuple !rxhash ...
	nic/wlan0/coalesce: rxc: ?,  txc: ?
	nic/wlan0/status:   down, speed=?, mtu=1500, txqlen=1000, gro_flush_timeout=0.000µs
	WARNING: nic/wlan0: TSO not enabled - TCP latency with packets > MSS will be poor

for rio:

	nic/eth0: Realtek Semiconductor Co., Ltd. RTL8111/8168/8411 PCI Express Gigabit Ethernet Controller rev 06
	nic/eth0/features: rx !tx !sg !tso !ufo !gso gro !lro rxvlan txvlan !ntuple !rxhash ...
	nic/eth0/coalesce: rxc: 200μs/4f/0μs-irq/0f-irq,  txc: 200μs/4f/0μs-irq/0f-irq
	nic/eth0/status:   up, speed=1000, mtu=1500, txqlen=1000, gro_flush_timeout=0.000µs
	WARNING: nic/eth0: TSO not enabled - TCP latency with packets > MSS will be poor
	WARNING: nic/eth0: RX coalesce latency is max 200μs - that will add to networked request-reply latency
	nic/eth1: Realtek Semiconductor Co., Ltd. RTL8111/8168/8411 PCI Express Gigabit Ethernet Controller rev 06
	nic/eth1/features: rx !tx !sg !tso !ufo !gso gro !lro rxvlan txvlan !ntuple !rxhash ...
	nic/eth1/coalesce: rxc: 0μs/1f/0μs-irq/0f-irq,  txc: 0μs/1f/0μs-irq/0f-irq
	nic/eth1/status:   down, speed=?, mtu=1500, txqlen=1000, gro_flush_timeout=0.000µs
	WARNING: nic/eth1: TSO not enabled - TCP latency with packets > MSS will be poor

The warning about "RX coalesce latency is max 200μs ..." says that on
receive path eth0 will be coalescing incoming frames for up to 200μs and
this way this delay will be added to overal latency. (for small frames
Realtek NICs do not coalesce interrupts - see details in the kernel patch).

Networked performance (raw and NEO) was not discussed in
http://navytux.spb.ru/~kirr/neo.html at all, but for the reference the
importance of C-states for performance was first found via this
networking latency benchmarks. Links on C-states topic:

	http://navytux.spb.ru/~kirr/neo.html#cpu-idle-c-states
	http://navytux.spb.ru/~kirr/neo.html#appendix-ii-cpu-c-states

Some draft history related to this patch:

	lab.nexedi.com/kirr/neo/commit/e8e395ae	X neotest: Move network benchmarking into separate function + add `neotest bench-net`
	lab.nexedi.com/kirr/neo/commit/a971231c	X neotest/info: Handle USB NICs
	lab.nexedi.com/kirr/neo/commit/5dd3d1ab	X neotest: sort NIC names
	lab.nexedi.com/kirr/neo/commit/9888f047	X neotest: Do not crash if kernel is too old to support gro_flush_timeout
	lab.nexedi.com/kirr/neo/commit/3a1bdf4a	X bench-remote / tcp : std benchmark output
	lab.nexedi.com/kirr/neo/commit/9450b6db	X bench-remote / ping += std bench output
	lab.nexedi.com/kirr/neo/commit/68d5b015	X show gro_flush_timeout + friends
	lab.nexedi.com/kirr/neo/commit/4c815af9	X neotest: Show NIC features and emit warning if !TSO
	lab.nexedi.com/kirr/neo/commit/659ce938	X neotest: Adjust ping and TCP RR sizes to fit 1 Ethernet frame, etc...
	lab.nexedi.com/kirr/neo/commit/ded384cb	X neotest += `lat_tcp.go -s`
	lab.nexedi.com/kirr/neo/commit/59d46504	X neotest += lat_tcp
	lab.nexedi.com/kirr/neo/commit/67fc3440	X show small (56B) and full-packet (1472B) ping link latencies
parent ee6c2796
...@@ -66,13 +66,23 @@ GOPATH=${GOPATH%:} ...@@ -66,13 +66,23 @@ GOPATH=${GOPATH%:}
# python # python
. $X/venv/bin/activate . $X/venv/bin/activate
# lmbench
export PATH=$X/lmbench/lmbench3/bin/`cd $X/lmbench/lmbench3/src; ../scripts/os`:$PATH
# ioping # ioping
export PATH=$X/ioping:$PATH export PATH=$X/ioping:$PATH
# XXX for mysqld # XXX for mysqld, ethtool
export PATH=$PATH:/sbin:/usr/sbin export PATH=$PATH:/sbin:/usr/sbin
EOF EOF
# NOTE lmbench before env.sh because env.sh uses `scripts/os` from lmbench
git clone -o kirr -b x/kirr https://lab.nexedi.com/kirr/lmbench.git
pushd lmbench/lmbench3/src
make -j`nproc`
go build -o ../bin/`../scripts/os`/lat_tcp_go lat_tcp.go
popd
. env.sh . env.sh
pip install pygolang # for tcpu.py pip install pygolang # for tcpu.py
...@@ -225,6 +235,11 @@ fkghz() { ...@@ -225,6 +235,11 @@ fkghz() {
python -c "print '%.2fGHz' % (`cat $1` / 1E6)" python -c "print '%.2fGHz' % (`cat $1` / 1E6)"
} }
# lspci1 <pcidev> <field> - show <field> from lspci information about <pcidev>
lspci1() {
lspci -vmm -s $1 |grep "^$2:\\s*" |sed -e "s/^$2:\\s*//"
}
# xhostname - show short system host name # xhostname - show short system host name
xhostname() { xhostname() {
# prefer first part of FQDN for misconfigured systems like # prefer first part of FQDN for misconfigured systems like
...@@ -378,6 +393,143 @@ system_info() { ...@@ -378,6 +393,143 @@ system_info() {
;; ;;
esac esac
# all NICs
# XXX warn if ethtool is not there
find /sys/class/net -type l -not -lname '*virtual*' |sort | \
while read nic; do
nicname=`basename $nic` # /sys/class/net/eth0 -> eth0
echo -n "nic/$nicname: "
nicdev=`realpath $nic/device` # /sys/class/net/eth0 -> /sys/devices/pci0000:00/0000:00:1f.6
case "$nicdev" in
*usb*)
# /sys/devices/pci0000:00/0000:00:1d.0/usb2/2-1/2-1.6/2-1.6:1.6 -> .../usb2/2-1/2-1.6
usbdev="$nicdev/.."
usbdev=`cat $usbdev/busnum`:`cat $usbdev/devnum` # ... -> 2:4
product=`lsusb -s $usbdev |sed -e 's/^Bus\s[0-9]*\sDevice\s[0-9]*:\sID\s[0-9a-f]*:[0-9a-f]*\s//'`
echo "$product (usb)"
;;
*pci*)
pcidev=`basename $nicdev` # /sys/devices/pci0000:00/0000:00:1f.6 -> 0000:00:1f.6
#lspci -s $pcidev
echo "`lspci1 $pcidev Vendor` `lspci1 $pcidev Device` rev `lspci1 $pcidev Rev`"
;;
*)
echo "$nicdev (TODO)"
;;
esac
nicwarnv=()
# show relevant features
featok=y
feat=`ethtool -k $nicname 2>/dev/null` || featok=n
if [ $featok != y ]; then
echo "nic/$nicname/features: ?"
else
# feat1 name abbrev -> abbrev. value (e.g. "tx" or "!tx")
feat1() {
# ntuple-filters: off [fixed]
v=`echo "$feat" |grep "^$1:\\s*" |awk '{print $2}'`
case $v in
on)
echo "$2"
;;
off)
echo "!$2"
;;
*)
echo "?($v)$2"
;;
esac
}
s="nic/$nicname/features:"
# NOTE feature abbrevs are those used by `ethtool -K` to set them
s+=" `feat1 rx-checksumming rx`"
s+=" `feat1 tx-checksumming tx`"
s+=" `feat1 scatter-gather sg`"
tso="`feat1 tcp-segmentation-offload tso`"
s+=" $tso"
s+=" `feat1 udp-fragmentation-offload ufo`"
s+=" `feat1 generic-segmentation-offload gso`"
s+=" `feat1 generic-receive-offload gro`"
s+=" `feat1 large-receive-offload lro`"
s+=" `feat1 rx-vlan-offload rxvlan`"
s+=" `feat1 tx-vlan-offload txvlan`"
s+=" `feat1 ntuple-filters ntuple`"
s+=" `feat1 receive-hashing rxhash`"
# ^^^ are the common features - others are specific to kernel/device
# XXX or list them all?
s+=" ..."
echo "$s"
# warn if !tso (linux starts enabling autocorking for e.g. small second segment and probably
# something else and lat_tcp latency grows stepwise from ~130μs to 500μs and more)
test "$tso" == "tso" || nicwarnv+=("TSO not enabled - TCP latency with packets > MSS will be poor")
fi
# show rx/tx coalescing latency
echo -n "nic/$nicname/coalesce:"
coalok=y
coal=`ethtool -c $nicname 2>/dev/null` || coalok=n
if [ $coalok != y ]; then
echo " rxc: ?, txc: ?"
else
# coal1 name -> value
coal1() {
echo "$coal" |grep "^$1:\\s*" | sed -e "s/^$1:\\s*//"
}
rxt=`coal1 rx-usecs`
rxf=`coal1 rx-frames`
rxt_irq=`coal1 rx-usecs-irq`
rxf_irq=`coal1 rx-frames-irq`
txt=`coal1 tx-usecs`
txf=`coal1 tx-frames`
txt_irq=`coal1 tx-usecs-irq`
txf_irq=`coal1 tx-frames-irq`
echo -en " rxc: ${rxt}μs/${rxf}f/${rxt_irq}μs-irq/${rxf_irq}f-irq,"
echo -e " txc: ${txt}μs/${txf}f/${txt_irq}μs-irq/${txf_irq}f-irq"
# XXX also add -low and -high ?
# warn if rx latency is too high
rxlat=$(($rxt>$rxt_irq?$rxt:$rxt_irq))
test "$rxlat" -le 10 || nicwarnv+=("RX coalesce latency is max ${rxlat}μs - that will add to networked request-reply latency")
fi
# show main parameters + GRO flush time
s="nic/$nicname/status: "
s+=" `cat $nic/operstate`"
speed=`cat $nic/speed 2>/dev/null` || speed=? # returns EINVAL for wifi
s+=", speed=$speed"
s+=", mtu=`cat $nic/mtu`"
s+=", txqlen=`cat $nic/tx_queue_len`"
if test -e $nic/gro_flush_timeout ; then
tgroflush_ns=`cat $nic/gro_flush_timeout`
s+=", gro_flush_timeout=`python -c "print '%.3f' % ($tgroflush_ns / 1E3)"`µs"
else
s+=", !gro_flush_timeout"
fi
echo "$s"
# XXX warn if gro_flush_timeout=0 or unsupported ?
# emit NIC warnings
for warn in "${nicwarnv[@]}"; do
echo "WARNING: nic/$nicname: $warn"
done
done
printf "%-20s" "sw/python:"; proginfo python --version 2>&1 # https://bugs.python.org/issue18338 printf "%-20s" "sw/python:"; proginfo python --version 2>&1 # https://bugs.python.org/issue18338
printf "%-20s" "sw/go:"; proginfo go version printf "%-20s" "sw/go:"; proginfo go version
printf "%-20s" "sw/sqlite:"; proginfo python -c \ printf "%-20s" "sw/sqlite:"; proginfo python -c \
...@@ -512,6 +664,101 @@ Benchmark$1-avg 1 \\3 \\4/op\ ...@@ -512,6 +664,101 @@ Benchmark$1-avg 1 \\3 \\4/op\
done done
} }
# hostof <url> - return hostname part of <url>
hostof() {
url=$1
python -c "import urlparse as p; u=p.urlparse(\"scheme://$url\"); print u.hostname"
}
# bench_net <url> - benchmark network
bench_net() {
url=$1
peer=`hostof $url`
shortpeer=`echo $peer |sed -e 's/\./ /' |awk '{print $1}'` # name.domain -> name
echo "node:"
echo -e "cluster:\t`xhostname`-$shortpeer"
echo -e "\n*** link latency:"
# ping2bench <topic> - converts timings from ping to std benchmark format
ping2bench() {
# rtt min/avg/max/mdev = 0.028/0.031/0.064/0.007 ms, ipg/ewma 0.038/0.032 ms
sed -u -e \
"s|^rtt min/avg/max/mdev = \([0-9.]\+\)/\([0-9.]\+\)/\([0-9.]\+\)/\([0-9.]\+\) ms.*\$|&\n\
Benchmark$1-min 1 \\1 ms/op\n\
Benchmark$1-avg 1 \\2 ms/op\
|"
}
# 16 = minimum ping payload size at which it starts putting struct timeval into payload and print RTT
# 1472 = 1500 (Ethernet MTU) - 20 (IPv4 header !options) - 8 (ICMPv4 header)
# 1452 = 1500 (Ethernet MTU) - 40 (IPv6 header !options) - 8 (ICMPv6 header)
# FIXME somehow IPv6 uses lower MTU than 1500 - recheck
sizev="16 1452" # max = min(IPv4, IPv6) so that it is always only 1 Ethernet frame on the wire
for size in $sizev; do
echo -e "\n# `xhostname`$peer (ping ${size}B)"
{ $profile sudo -n ping -i0 -w 3 -s $size -q $peer || \
echo "# skipped -> enable ping in sudo for `whoami`@`xhostname`"; } | \
ping2bench pingrtt-/${size}B
echo -e "\n# $peer`xhostname` (ping ${size}B)"
# TODO profile remotely
on $url "sudo -n ping -i0 -w3 -s ${size} -q \$(echo \${SSH_CONNECTION%% *}) || \
echo \\\"# skipped -> enable ping in sudo for \`whoami\`@${peer}\\\"" | \
ping2bench -pingrtt/${size}B
done
# TODO
# echo 1 > /proc/sys/net/ipv4/tcp_low_latency
# netstat -s
# /sys/class/net/ethX/gro_flush_timeout
# /proc/sys/net/ipv4/tcp_limit_output_bytes
# ( https://lwn.net/Articles/507065/ "The default value of this
# limit is 128KB; it could be set lower on systems where latency is the primary concern" )
# ? tcp pacing
# net.ipv4.tcp_autocorking (f54b3111 "tcp: auto corking")
echo -e "\n*** TCP latency:"
# lattcp2bench <topic> - convert timings from lat_tcp to std benchmark format
lattcp2bench() {
# TCP latency using neo2: 52.3468 microseconds
sed -u -e \
"s|^TCP latency using .*: \([0-9.]\+\) microseconds.*\$|Benchmark$1 1 \\1 µs/op\t# &|"
}
# 1 = minimum TCP payload
# 1460 = 1500 (Ethernet MTU) - 20 (IPv4 header !options) - 20 (TCP header !options)
# 1440 = 1500 (Ethernet MTU) - 40 (IPv6 header !options) - 20 (TCP header !options)
# FIXME somehow IPv6 uses lower MTU than 1500 - recheck
sizev="1 1400 1500 4096" # 1400 = 1440 - ε (1 eth frame); 1500 = 1440 + ε (2 eth frames); 4096 - just big 4K (3 eth frames)
for size in $sizev; do
echo -e "\n# `xhostname`$peer (lat_tcp.c ${size}B -> lat_tcp.c -s)"
# TODO profile remotely
on $url "nohup lat_tcp -s </dev/null >/dev/null 2>/dev/null &"
nrun lat_tcp -m $size $peer | lattcp2bench "tcprtt(c_c)-/${size}B"
lat_tcp -S $peer
echo -e "\n# `xhostname`$peer (lat_tcp.c ${size}B -> lat_tcp.go -s)"
# TODO profile remotely
on $url "nohup lat_tcp_go -s </dev/null >/dev/null 2>/dev/null &"
nrun lat_tcp -m $size $peer | lattcp2bench "tcprtt(c_go)-/${size}B"
lat_tcp -S $peer
echo -e "\n# $peer`xhostname` (lat_tcp.c ${size}B -> lat_tcp.c -s)"
lat_tcp -s
# TODO profile remotely
nrun on $url "lat_tcp -m $size \${SSH_CONNECTION%% *}" | lattcp2bench "-tcprtt(c_c)/${size}B"
lat_tcp -S localhost
echo -e "\n# $peer`xhostname` (lat_tcp.c ${size}B -> lat_tcp.go -s)"
lat_tcp_go -s 2>/dev/null &
# TODO profile remotely
nrun on $url "lat_tcp -m $size \${SSH_CONNECTION%% *}" | lattcp2bench "-tcprtt(c_go)/${size}B"
lat_tcp -S localhost
done
}
# command: benchmark local disk # command: benchmark local disk
...@@ -524,6 +771,13 @@ cmd_bench-cpu() { ...@@ -524,6 +771,13 @@ cmd_bench-cpu() {
bench_cpu bench_cpu
} }
# command: benchmark network
cmd_bench-net() {
url=$1
test -z "$url" && die "Usage: neotest bench-net [user@]<host>:<path>"
bench_net $url
}
# command: print information about local node # command: print information about local node
cmd_info-local() { cmd_info-local() {
system_info system_info
...@@ -562,6 +816,7 @@ The commands are: ...@@ -562,6 +816,7 @@ The commands are:
bench-cpu benchmark local cpu bench-cpu benchmark local cpu
bench-disk benchmark local disk bench-disk benchmark local disk
bench-net benchmark network
deploy deploy NEO & needed software for tests to remote host deploy deploy NEO & needed software for tests to remote host
...@@ -589,6 +844,7 @@ test-py) f=( );; ...@@ -589,6 +844,7 @@ test-py) f=( );;
bench-cpu) f=(build );; bench-cpu) f=(build );;
bench-disk) f=( fs );; bench-disk) f=( fs );;
bench-net) f=( net );;
info) f=( );; info) f=( );;
info-local) f=( net );; info-local) f=( net );;
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment