Enhance iperf3 to run for multiple sessions by Suneeth-D · Pull Request #551 · intel/lkp-tests

Suneeth-D · 2026-04-10T10:25:11Z

This patch series on the iperf3 microbenchmark package focuses on enhancing it to have few functionalities mentioned below and address the consequences of having those while not disturbing the existing iperf3 package on the latest intel/lkp-tests: master.

The enhancements are as follows:-

Modify PKGBUILD to have iperf3 installation from the upstream
repo as a single source rather than picking up distro specific which is
non-uniform among the distros.
Add sessions and port params to have multiple sessions of iperf3
running as client-server pairs which helps in identifying regression
on multi-core systems.
Add numactl param to enable running iperf3 in specified memory
policy on systems whers NR_NUMA_NODES > 1.
Add tune param to have certain kernel parameters pertaining to
network stack tuned before running the iperf3 microbenchmark.
Add unittests for iperf covering both single and multiple sessions.

Below is the yaml file I used for testing my patch series which has all the additional params that is introduced.

iperf3.yaml

#! jobs/iperf.yaml
suite: iperf
testcase: iperf
category: benchmark
runtime: 30s
sessions: 8
port: 25000
cluster: cs-localhost
if role client:
iperf:
protocol: tcp
numactl: true
tune:
- net.core.wmem_max 212992
- net.core.rmem_max 212992
- net.core.rmem_default 212992
- net.core.wmem_default 212992
- net.core.netdev_max_backlog 25000
- net.core.busy_poll 30
- net.core.busy_read 30
- net.core.netdev_budget 600
- net.core.rps_sock_flow_entries 65536
job_origin: jobs/iperf.yaml
arch: x86_64
node_roles: server client

UNIT TEST output after adding the patch series,

rake spec spec=stats

PWD = /home/suneeth/lkp-tests
ENV['LKP_SRC'] = /home/suneeth/lkp-tests
/usr/bin/ruby3.2 -I/var/lib/gems/3.2.0/gems/rspec-core-3.13.6/lib:/var/lib/gems/3.2.0/gems/rspec-support-3.13.7/lib /var/lib/gems/3.2.0/gems/rspec-core-3.13.6/exe/rspec --pattern spec/**{,/*/**}/stats_spec.rb
..........................................................................................................................................................................................................................................................................................................................................................................................F.....................................................

Failures:

stats scripts invariance: /home/suneeth/lkp-tests/spec/fixtures/stats/vmstat/1
Failure/Error: expect(new_stat).to eq old_stat.chomp

expected: "time: 1420978715\nprocs.r: 2\nprocs.b: 0\nmemory.swpd: 0\nmemory.free: 224636\nmemory.buff: 33636\nm...\nio.bo: 4\nsystem.in: 753\nsystem.cs: 2187\ncpu.us: 2\ncpu.sy: 1\ncpu.id: 98\ncpu.wa: 0\ncpu.st: 0"
got: "time: 1421007515\nprocs.r: 2\nprocs.b: 0\nmemory.swpd: 0\nmemory.free: 224636\nmemory.buff: 33636\nm...\nio.bo: 4\nsystem.in: 753\nsystem.cs: 2187\ncpu.us: 2\ncpu.sy: 1\ncpu.id: 98\ncpu.wa: 0\ncpu.st: 0"

(compared using ==)

Diff:
@@ -1,4 +1,4 @@
-time: 1420978715
+time: 1421007515
procs.r: 2
procs.b: 0
memory.swpd: 0
@@ -16,7 +16,7 @@
cpu.id: 98
cpu.wa: 0
cpu.st: 0
-time: 1420978716
+time: 1421007516
procs.r: 0
procs.b: 0
memory.swpd: 0

./spec/unit/lib/stats_spec.rb:26:in `block (4 levels) in <top (required)>'

Finished in 1 minute 58.13 seconds (files took 0.33267 seconds to load)
432 examples, 1 failure

Failed examples:

rspec ./spec/unit/lib/stats_spec.rb[1:1:379] # stats scripts invariance: /home/suneeth/lkp-tests/spec/fixtures/stats/vmstat/1

/usr/bin/ruby3.2 -I/var/lib/gems/3.2.0/gems/rspec-core-3.13.6/lib:/var/lib/gems/3.2.0/gems/rspec-support-3.13.7/lib /var/lib/gems/3.2.0/gems/rspec-core-3.13.6/exe/rspec --pattern spec/**{,/*/**}/stats_spec.rb failed

NOTE: The failure on vmstat unit testcase is already there on latest master (`f182b61`)

rli9 · 2026-04-14T01:31:44Z

-{
-	threads_to_iterate=1
-	nr_node=$(echo /sys/devices/system/node/node* | wc -w)
+threads_to_iterate=1


thanks, any consideration to move these variables out of the setup_threads_to_iterate()?

My intention was to expose the value of cores_per_node which was local to setup_threads_to_iterate(). But thanks to you, since you pointed this out, I was thinking maybe I can call keep all the variables inside the setup_threads_to_iterate(), export cores_per_node and call this function once in the run script to get and use the value. That way the implementation seems to be cleaner and safer I guess.

rli9 · 2026-04-14T01:32:13Z

  runtime:
  protocol:
+  sessions:
+  port:


is port necessary to be part of the result root path?

Yes you are right. It makes sense to remove -port from the result-root path. Will make the changes. Thanks much for the suggestion.

rli9 · 2026-04-14T01:34:59Z

+        val=$2
+        shift 2
+
+        [ -n "$key" ] && [ -n "$val" ] || continue


consider to use double brackets like [[, ]]

Acknowledged. Thanks for catching that.

rli9 · 2026-04-14T01:35:02Z

+tune_param()
+{
+    set -- $tune
+    while [ $# -ge 2 ]; do


consider to use double brackets like [[, ]]

Acknowledged. Thanks again.

rli9 · 2026-04-14T01:39:38Z

+
+build()
+{
+	cd $srcdir/$pkgname


consider to use cd_src_pkg_dir by source $LKP_SRC/lib/tests/pkgbuild.sh

Acknowledged. Thanks for the suggestion.

Copilot

Pull request overview

This PR enhances the iperf benchmark integration to support multi-session runs, additional runtime tuning options, and builds iperf from upstream sources for more consistent cross-distro behavior.

Changes:

Add multi-session execution support with configurable sessions/port, optional numactl, and sysctl tuning in programs/iperf/run.
Update iperf stats parsing to handle concatenated JSON from multiple sessions and report averaged throughput.
Introduce a makepkg-based upstream PKGBUILD and add new fixtures to unit-test iperf stats parsing for TCP/UDP single- and multi-session outputs.

Reviewed changes

Copilot reviewed 16 out of 16 changed files in this pull request and generated 9 comments.

Show a summary per file

File	Description
`programs/iperf/run`	Adds sessions/port/numactl/tune handling and multi-session orchestration.
`programs/iperf/parse`	Parses concatenated multi-session JSON and averages bps metrics.
`programs/iperf/meta.yaml`	Exposes new `sessions`, `port`, `numactl`, `tune` parameters.
`programs/iperf/pkg/PKGBUILD`	Adds upstream-source build packaging for iperf via makepkg.
`programs/iperf/pkg/depends`	Removes prior runtime dependency declaration.
`lib/sysinfo.sh`	Changes how NUMA/core topology vars are computed/exported.
`distro/adaptation-pkg/ubuntu`	Adds `iperf` to makepkg adaptation list.
`distro/adaptation-pkg/debian`	Adds `iperf` to makepkg adaptation list.
`spec/fixtures/stats/iperf/tcp`	Adds iperf3 TCP JSON fixture for stats invariance tests.
`spec/fixtures/stats/iperf/tcp.yaml`	Adds expected TCP stats output fixture.
`spec/fixtures/stats/iperf/tcp-multi-session`	Adds multi-session TCP concatenated JSON fixture.
`spec/fixtures/stats/iperf/tcp-multi-session.yaml`	Adds expected multi-session TCP stats output fixture.
`spec/fixtures/stats/iperf/udp`	Adds iperf3 UDP JSON fixture for stats invariance tests.
`spec/fixtures/stats/iperf/udp.yaml`	Adds expected UDP stats output fixture.
`spec/fixtures/stats/iperf/udp-multi-session`	Adds multi-session UDP concatenated JSON fixture.
`spec/fixtures/stats/iperf/udp-multi-session.yaml`	Adds expected multi-session UDP stats output fixture.

Comments suppressed due to low confidence (1)

programs/iperf/pkg/depends:1

programs/iperf/run (and programs/iperf-server/daemon) invoke iperf3, but programs/iperf/pkg/depends is now empty. On a clean rootfs this removes the only declared dependency that ensures iperf3 is present, and it can also cause sbin/makepkg-deps/build_depends_pkg to fail with "empty deps for iperf". Add the appropriate dependency declaration (either restore a distro dependency like iperf3, or list the makepkg package(s) that provide iperf3).

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-04-14T01:50:20Z

+	for i in `seq 1 $sessions`; do
+		if [ "$numactl" = true ]; then
+			numactl_server="numactl -C $(($cores_per_node + i))"
+		fi
+		$numactl_server iperf3 -s -B $server -p $(($port+i-1)) -1 > /dev/null &
+	done
+
+	for i in `seq 1 $sessions`; do
+		if [ "$numactl" = true ]; then
+			numactl_client="numactl -C $((0 + i))"
+		fi
+		$numactl_server iperf3 -c $server -p $(($port+i-1)) -i 10 -T $((0+i-1)) -J -t $runtime $opt_udp
+	done


The "multi-session" mode currently runs the client iperf3 commands sequentially (no &/wait). That means only one session is active at a time, which defeats the stated goal of running multiple client/server pairs concurrently to stress multiple cores. If true concurrency is required, start the client sessions in the background, capture each session’s JSON output to separate files, then wait and concatenate them in a deterministic order for parsing.

Suggested change

for i in `seq 1 $sessions`; do

if [ "$numactl" = true ]; then

numactl_server="numactl -C $(($cores_per_node + i))"

fi

$numactl_server iperf3 -s -B $server -p $(($port+i-1)) -1 > /dev/null &

done

for i in `seq 1 $sessions`; do

if [ "$numactl" = true ]; then

numactl_client="numactl -C $((0 + i))"

fi

$numactl_server iperf3 -c $server -p $(($port+i-1)) -i 10 -T $((0+i-1)) -J -t $runtime $opt_udp

done

client_output_dir=$(mktemp -d)

declare -a server_pids

declare -a client_pids

declare -a client_output_files

client_cmd=""

for i in `seq 1 $sessions`; do

if [ "$numactl" = true ]; then

numactl_server="numactl -C $(($cores_per_node + i))"

else

numactl_server=""

fi

$numactl_server iperf3 -s -B $server -p $(($port+i-1)) -1 > /dev/null &

server_pids[$i]=$!

done

for i in `seq 1 $sessions`; do

if [ "$numactl" = true ]; then

numactl_client="numactl -C $((0 + i))"

else

numactl_client=""

fi

client_output_file="$client_output_dir/client-$i.json"

client_output_files[$i]="$client_output_file"

client_session_cmd="$numactl_client iperf3 -c $server -p $(($port+i-1)) -i 10 -T $((0+i-1)) -J -t $runtime $opt_udp"

if [ -n "$client_cmd" ]; then

client_cmd="${client_cmd}; "

fi

client_cmd="${client_cmd}${client_session_cmd}"

eval "$client_session_cmd > \"$client_output_file\" &"

client_pids[$i]=$!

done

for i in `seq 1 $sessions`; do

wait "${client_pids[$i]}"

done

for i in `seq 1 $sessions`; do

wait "${server_pids[$i]}"

done

for i in `seq 1 $sessions`; do

cat "${client_output_files[$i]}"

done

rm -rf "$client_output_dir"

Copilot · 2026-04-14T01:50:20Z

+	for i in `seq 1 $sessions`; do
+		if [ "$numactl" = true ]; then
+			numactl_server="numactl -C $(($cores_per_node + i))"
+		fi
+		$numactl_server iperf3 -s -B $server -p $(($port+i-1)) -1 > /dev/null &
+	done


Servers are started in the background and clients are started immediately afterward, but there is no readiness check to ensure each iperf3 -s is actually listening on its port before the corresponding client connects. This can cause intermittent connection failures (and leave iperf3 -s -1 processes lingering, waiting forever). Add a small per-port readiness loop (e.g., ss/lsof/nc) and/or a cleanup trap to kill spawned servers on failure.

Copilot · 2026-04-14T01:50:21Z

+threads_to_iterate=1
+export nr_node=$(echo /sys/devices/system/node/node* | wc -w)

-	local threads_per_core
-	local cores_per_node
-	local cores_per_llc
-	local nr_l3
-	local nr_l2
-	threads_per_core=$(lscpu | awk '/Thread\(s\) per core:/ { print $4 }')
-	nr_l3=$(lscpu | grep 'L3' | awk -F '[()]' '{print $2}' | awk '{print $1}')
-	nr_l2=$(lscpu | grep 'L2' | awk -F '[()]' '{print $2}' | awk '{print $1}')
-	cores_per_llc=$((nr_l2 / nr_l3))
-	cores_per_node=$((nr_cpu / nr_node / threads_per_core))
+threads_per_core=$(lscpu | awk '/Thread\(s\) per core:/ { print $4 }')
+nr_l3=$(lscpu | grep 'L3' | awk -F '[()]' '{print $2}' | awk '{print $1}')
+nr_l2=$(lscpu | grep 'L2' | awk -F '[()]' '{print $2}' | awk '{print $1}')
+cores_per_llc=$((nr_l2 / nr_l3))
+export cores_per_node=$((nr_cpu / nr_node / threads_per_core))


lib/sysinfo.sh now computes and exports cores_per_node at file load time using nr_cpu, but nr_cpu is not defined within this script. If the script is sourced before nr_cpu is exported by the harness, the arithmetic expansion will fail and can break callers. Consider moving this computation back inside setup_threads_to_iterate() (or guarding it with a check/default for nr_cpu).

Copilot · 2026-04-14T01:50:21Z

+if [[ -n "$sessions" ]];then
+	for i in `seq 1 $sessions`; do
+		if [ "$numactl" = true ]; then
+			numactl_server="numactl -C $(($cores_per_node + i))"
+		fi
+		$numactl_server iperf3 -s -B $server -p $(($port+i-1)) -1 > /dev/null &
+	done
+
+	for i in `seq 1 $sessions`; do
+		if [ "$numactl" = true ]; then
+			numactl_client="numactl -C $((0 + i))"
+		fi
+		$numactl_server iperf3 -c $server -p $(($port+i-1)) -i 10 -T $((0+i-1)) -J -t $runtime $opt_udp


When sessions is set, this script starts iperf3 -s processes locally and then runs the clients against $server. If $server is a remote host (e.g. direct_server_ips is set), the local servers are unnecessary and -B $server will try to bind to an address not present on the client, causing failures. Consider restricting the local server spawning to localhost-only runs (or moving server spawning to the server-role program) and keeping the client-only behavior for remote server cases.

Suggested change

if [[ -n "$sessions" ]];then

for i in `seq 1 $sessions`; do

if [ "$numactl" = true ]; then

numactl_server="numactl -C $(($cores_per_node + i))"

fi

$numactl_server iperf3 -s -B $server -p $(($port+i-1)) -1 > /dev/null &

done

for i in `seq 1 $sessions`; do

if [ "$numactl" = true ]; then

numactl_client="numactl -C $((0 + i))"

fi

$numactl_server iperf3 -c $server -p $(($port+i-1)) -i 10 -T $((0+i-1)) -J -t $runtime $opt_udp

server_is_localhost=false

case "$server" in

127.0.0.1|::1|localhost)

server_is_localhost=true

;;

esac

if [[ -n "$sessions" ]];then

if [ "$server_is_localhost" = true ]; then

for i in `seq 1 $sessions`; do

if [ "$numactl" = true ]; then

numactl_server="numactl -C $(($cores_per_node + i))"

fi

$numactl_server iperf3 -s -B $server -p $(($port+i-1)) -1 > /dev/null &

done

fi

for i in `seq 1 $sessions`; do

if [ "$numactl" = true ]; then

numactl_client="numactl -C $((0 + i))"

fi

$numactl_client iperf3 -c $server -p $(($port+i-1)) -i 10 -T $((0+i-1)) -J -t $runtime $opt_udp

Copilot · 2026-04-14T01:50:21Z

+		if [ "$numactl" = true ]; then
+			numactl_client="numactl -C $((0 + i))"
+		fi
+		$numactl_server iperf3 -c $server -p $(($port+i-1)) -i 10 -T $((0+i-1)) -J -t $runtime $opt_udp


In the multi-session client loop you build a numactl_client command, but the actual iperf3 invocation uses numactl_server again. This makes the client sessions run with the server CPU binding (and numactl_client is effectively unused). Switch the client invocation to use numactl_client (or a shared variable) so client/server bindings behave as intended.

Suggested change

$numactl_server iperf3 -c $server -p $(($port+i-1)) -i 10 -T $((0+i-1)) -J -t $runtime $opt_udp

$numactl_client iperf3 -c $server -p $(($port+i-1)) -i 10 -T $((0+i-1)) -J -t $runtime $opt_udp

Copilot · 2026-04-14T01:50:21Z

+	client_cmd="iperf3 -t $runtime -J -c $server $opt_udp"
+	${client_cmd}
+fi

 echo "$(date +'%F %T') ${client_cmd}" >> $RESULT_ROOT/reproduce.sh


client_cmd is only set in the single-session branch, but it is referenced unconditionally when appending to $RESULT_ROOT/reproduce.sh. When sessions is set this writes an empty/incorrect reproduce command. Define an appropriate reproduce command for the multi-session case (or gate the reproduce.sh write on client_cmd being set).

Copilot · 2026-04-14T01:50:21Z

+	make install
+}
+
+package()
+{
+	 benchmark_path="${pkgdir}/lkp/benchmarks/${pkgname}"
+	 mkdir -p $benchmark_path
+	 cp -r "$srcdir/$pkgname/" "$benchmark_path/".


This PKGBUILD runs make install during build(), which will install into the build host rather than staging into $pkgdir. For reproducible packaging (and to avoid polluting the build environment), keep build() to compilation only and perform installation in package() using DESTDIR="$pkgdir"/--prefix, packaging just the installed artifacts rather than copying the whole source tree.

Suggested change

make install

}

package()

{

benchmark_path="${pkgdir}/lkp/benchmarks/${pkgname}"

mkdir -p $benchmark_path

cp -r "$srcdir/$pkgname/" "$benchmark_path/".

}

package()

{

cd $srcdir/$pkgname

make DESTDIR="$pkgdir" install

Copilot · 2026-04-14T01:50:22Z

+pkgver=git
+pkgrel=1
+url='https://github.com/esnet/iperf'
+arch=('i386' 'x86_64' 'aarch64')


The arch list uses i386, but other repo PKGBUILDs use Arch-style i686 (e.g., programs/kernel-selftests/pkg/PKGBUILD:4). If this is built via makepkg, i386 may be treated as invalid/unsupported. Align the architecture identifiers with the existing PKGBUILD convention in this repo.

Suggested change

arch=('i386' 'x86_64' 'aarch64')

arch=('i686' 'x86_64' 'aarch64')

Copilot · 2026-04-14T01:50:22Z

+		if [ "$numactl" = true ]; then
+			numactl_server="numactl -C $(($cores_per_node + i))"
+		fi
+		$numactl_server iperf3 -s -B $server -p $(($port+i-1)) -1 > /dev/null &


The new numactl handling here uses numactl -C ... (CPU affinity) only. That does not apply any NUMA memory policy (e.g., --membind/--interleave) and also bypasses the repo’s existing NUMA binding helpers (lib/numactl.sh, used by e.g. programs/memtier/run:26-83). If the intent is memory policy control, consider wiring this to parse_numa_node_binding/numa_node_binding so both CPU+memory bindings are handled consistently.

rli9 · 2026-04-21T02:15:58Z

Thanks, @Suneeth-D could you help do a check of "Ruby / test-code (3.2) (pull_request)Failing after 1m"?

Get iperf3 microbenchmark package from a reliable source rather than using distro specific version of iperf which varies from distro to distro. Signed-off-by: Suneeth D <Suneeth.D@amd.com>

The patch has the following changes incorporated:- * Add sessions param to enable run multiple sessions of iperf3 server-client pairs. * In order to avoid server unavailability and port conflicts, add port param to run $sessions sessions of iperf3 server-client pairs on ports $port through $port+($sessions-1). * Modify the existing parser script, thereby enabling parse the "bps" for multiple iperf3 sessions Signed-off-by: Suneeth D <Suneeth.D@amd.com>

Add numactl param to enable iperf3 when run in multiple server-client pairs, with a specific memory placement policy. Signed-off-by: Suneeth D <Suneeth.D@amd.com>

Add tune param to run script and meta.yaml to pass the kernel parameters that needs to be tuned wrt networking before starting to run the iperf3 microbenchmark Signed-off-by: Suneeth D <Suneeth.D@amd.com>

Add appropiate yaml files and console logs for tcp/udp(single/multiple) sessions under spec/fixtures/stats dir for the iperf to have unit tests coverage for multiple sessions. Signed-off-by: Suneeth D <Suneeth.D@amd.com>

rli9 · 2026-04-21T08:10:19Z

Thanks, some remaining issue to help check at
Ruby / test-code (3.1) (pull_request).. Forget to mention, you can run tools/sort-files to sort the file contents.

rli9 · 2026-04-21T10:29:39Z

thanks for the patch

rli9 reviewed Apr 14, 2026

View reviewed changes

rli9 requested a review from Copilot April 14, 2026 01:40

Copilot started reviewing on behalf of rli9 April 14, 2026 01:40 View session

Copilot AI reviewed Apr 14, 2026

View reviewed changes

Suneeth-D force-pushed the iperf_enhance branch from 8def7b2 to 1a33c83 Compare April 16, 2026 08:28

Suneeth-D requested a review from rli9 April 20, 2026 20:07

Suneeth-D force-pushed the iperf_enhance branch from 1a33c83 to 0ba6c52 Compare April 21, 2026 05:50

Suneeth-D added 5 commits April 21, 2026 07:48

Add PKGBUILD script for iperf installation

0ad0f64

Get iperf3 microbenchmark package from a reliable source rather than using distro specific version of iperf which varies from distro to distro. Signed-off-by: Suneeth D <Suneeth.D@amd.com>

Add numactl param

ef75156

Add numactl param to enable iperf3 when run in multiple server-client pairs, with a specific memory placement policy. Signed-off-by: Suneeth D <Suneeth.D@amd.com>

Add tune param

d60836c

Add tune param to run script and meta.yaml to pass the kernel parameters that needs to be tuned wrt networking before starting to run the iperf3 microbenchmark Signed-off-by: Suneeth D <Suneeth.D@amd.com>

Extend iperf unit test for multiple sessions

b504de3

Add appropiate yaml files and console logs for tcp/udp(single/multiple) sessions under spec/fixtures/stats dir for the iperf to have unit tests coverage for multiple sessions. Signed-off-by: Suneeth D <Suneeth.D@amd.com>

Suneeth-D force-pushed the iperf_enhance branch from 0ba6c52 to b504de3 Compare April 21, 2026 09:33

rli9 merged commit fbdf70c into intel:master Apr 21, 2026
12 checks passed

-	for i in `seq 1 $sessions`; do
-		if [ "$numactl" = true ]; then
-			numactl_server="numactl -C $(($cores_per_node + i))"
-		fi
-		$numactl_server iperf3 -s -B $server -p $(($port+i-1)) -1 > /dev/null &
-	done
-	for i in `seq 1 $sessions`; do
-		if [ "$numactl" = true ]; then
-			numactl_client="numactl -C $((0 + i))"
-		fi
-		$numactl_server iperf3 -c $server -p $(($port+i-1)) -i 10 -T $((0+i-1)) -J -t $runtime $opt_udp
-	done
+	client_output_dir=$(mktemp -d)
+	declare -a server_pids
+	declare -a client_pids
+	declare -a client_output_files
+	client_cmd=""
+	for i in `seq 1 $sessions`; do
+		if [ "$numactl" = true ]; then
+			numactl_server="numactl -C $(($cores_per_node + i))"
+		else
+			numactl_server=""
+		fi
+		$numactl_server iperf3 -s -B $server -p $(($port+i-1)) -1 > /dev/null &
+		server_pids[$i]=$!
+	done
+	for i in `seq 1 $sessions`; do
+		if [ "$numactl" = true ]; then
+			numactl_client="numactl -C $((0 + i))"
+		else
+			numactl_client=""
+		fi
+		client_output_file="$client_output_dir/client-$i.json"
+		client_output_files[$i]="$client_output_file"
+		client_session_cmd="$numactl_client iperf3 -c $server -p $(($port+i-1)) -i 10 -T $((0+i-1)) -J -t $runtime $opt_udp"
+		if [ -n "$client_cmd" ]; then
+			client_cmd="${client_cmd}; "
+		fi
+		client_cmd="${client_cmd}${client_session_cmd}"
+		eval "$client_session_cmd > \"$client_output_file\" &"
+		client_pids[$i]=$!
+	done
+	for i in `seq 1 $sessions`; do
+		wait "${client_pids[$i]}"
+	done
+	for i in `seq 1 $sessions`; do
+		wait "${server_pids[$i]}"
+	done
+	for i in `seq 1 $sessions`; do
+		cat "${client_output_files[$i]}"
+	done
+	rm -rf "$client_output_dir"

	$numactl_server iperf3 -c $server -p $(($port+i-1)) -i 10 -T $((0+i-1)) -J -t $runtime $opt_udp
	$numactl_client iperf3 -c $server -p $(($port+i-1)) -i 10 -T $((0+i-1)) -J -t $runtime $opt_udp

	arch=('i386' 'x86_64' 'aarch64')
	arch=('i686' 'x86_64' 'aarch64')

Uh oh!

Conversation

Suneeth-D commented Apr 10, 2026

The enhancements are as follows:-

iperf3.yaml

UNIT TEST output after adding the patch series,

rake spec spec=stats

./spec/unit/lib/stats_spec.rb:26:in `block (4 levels) in <top (required)>'

NOTE: The failure on vmstat unit testcase is already there on latest master (f182b61)

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Suneeth-D Apr 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Apr 14, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 14, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 14, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 14, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 14, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 14, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 14, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 14, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 14, 2026

Choose a reason for hiding this comment

Uh oh!

rli9 commented Apr 21, 2026

Uh oh!

rli9 commented Apr 21, 2026

Uh oh!

Uh oh!

rli9 commented Apr 21, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

NOTE: The failure on vmstat unit testcase is already there on latest master (`f182b61`)

Suneeth-D Apr 16, 2026 •

edited

Loading