Docs

Elastic Compute Service

Performance test

Testing Nginx performance

To learn the network performance of your ECS instance, you may typically analyze requests completed per second of the Nginx service (for long and short connections). This topic describes a test plan to obtain the optimal Nginx performance of ECS instances.

Testing tools and versions

In this topic, the Nginx service is deployed on the Server, and the Wrk pressure testing tool is deployed on the Client.

Nginx: A high-performance, lightweight web server for HTTP services and reverse proxy. This topic uses Nginx 1.18.0 as an example.

Wrk: A lightweight HTTP benchmarking tool. This topic uses wrk [epoll] Copyright (C) 2012 Will Glozer as an example.

Test metric

Requests/sec: The average number of requests completed per second. The higher this value, the better the Nginx performance.

Test environment

Two ECS instances with the same specifications are used, which serve as the service side (Server) and client side (Client), respectively. The specific configuration is shown in the table below.

To create an instance, refer to Purchasing an instance by following the wizard.

Sample configuration	Server	Client
Instance type	ecs.g3i.2xlarge	ecs.g3i.2xlarge
Image type	Ubuntu 22.04	Ubuntu 22.04
Instance quantity	1	1

Procedure

Step 1: Deploy and start the Nginx service on the Server

Deploy the Nginx service.

Execute the following command to create an Nginx deployment file.

vim nginx_install.sh

Press i key to enter edit mode, and then enter the following content.

#!/bin/bash
# install nginx
if which yum; then
rpm -Uvh https://dl.fedoraproject.org/pub/epel/epel-release-latest-7.noarch.rpm
yum -y install nginx
elif which apt; then
apt update
apt -y install nginx
fi

# close accesslog
mv /etc/nginx/nginx.conf /etc/nginx/nginx.conf.origin
sed "/ *access_log.*/c\access_log off;" /etc/nginx/nginx.conf.origin > /etc/nginx/nginx.conf

# prepare demo file
html_dir=$(grep root /etc/nginx/nginx.conf |grep -v "#"|awk '{print $2}'|sed "s/;//")
if [ ! ${html_dir} ];then
html_dir=$(grep root /etc/nginx/sites-available/default|grep -v "#"|awk '{print $2}'|sed "s/;//")
fi
echo "nginx html dir is ${html_dir}"

cd ${html_dir}
dd if=/dev/urandom of=1k bs=1k count=1
dd if=/dev/urandom of=4k bs=4k count=1
dd if=/dev/urandom of=1M bs=1M count=1
echo "hello">hello

echo "install and prepare nginx finish"
nginx -v

Press the Esc key to exit edit mode, enter :wq, and then press Enter to save and exit the file.

Execute the following command to install Nginx.

bash nginx_install.sh

Start the Nginx service on the Server.

Execute the following command to create a run file "nginx_run.sh".

vim nginx_run.sh

Press i key to enter edit mode, and then enter the following content.

#!/bin/bash
# run
# change ulimit
ulimit -n 65535
ulimit -n
# start nginx service
systemctl stop nginx
systemctl start nginx
systemctl status nginx
echo "start nginx finish"

Execute the following command to start the Nginx service.

bash nginx_run.sh

Step 2: Optimize instance configurations for optimal Nginx performance (optional)

For detailed items and steps, refer to Optimal configuration.

Step 3: Deploy Wrk on the Client and create a test script

Deploy Wrk.

Execute the following command to create a Wrk deployment file.

vim wrk_install.sh

Press i key to enter edit mode, and then enter the following content.

#!/bin/bash

if which wrk; then echo 'wrk is already installed' && exit; fi

if which yum;then yum install -y unzip python3; elif which apt ; then apt update && apt -y install unzip python3; fi

cd $(dirname $0)
#install wrk
while true
do
wget https://github.com/wg/wrk/archive/refs/heads/master.zip && break
done
unzip master.zip
cd wrk-master/
make
cp wrk /usr/bin/
whereis wrk

Press the Esc key to exit edit mode, enter :wq, and then press Enter to save and exit the file.

Execute the following command to install Wrk.

bash wrk_install.sh

Create the test script wrk_run.sh.

Execute the following command to create the test script.

vim wrk_run.sh

Press i key to enter edit mode, and then enter the following content.

#!/bin/bash

ulimit -n 65535

if [ $# -lt 1 ]; then
echo 'Usage: bash wrk_run.sh $serverIP [$threadNums] [$connNums]'
exit
elif [ $# -gt 3 ]; then
echo 'Usage: bash wrk_run.sh $serverIP [$threadNums] [$connNums]'
echo "Error: The number of parameters must be between 1-3"
exit
fi

dest_host=$1
if [ $# -eq 2 ]; then
wrk_thread=$2
elif [ $# -eq 3 ]; then
wrk_thread=$2
wrk_conn=$3
fi

if [ ! ${wrk_thread} ];then wrk_thread=128; fi
if [ ! ${wrk_conn} ];then wrk_conn=1024; fi

for data in hello
do

# long connection
echo "$(date) BENCHMARK STATR to stress nginx_long_ops[$data] with wrk ${dest_host} ${wrk_thread} ${wrk_conn} "
wrk -t ${wrk_thread} -c ${wrk_conn} -d 60 http://${dest_host}/$data
echo ""

# short connection
echo "$(date) BENCHMARK STATR to stress nginx_short_ops[$data] with wrk ${dest_host} ${wrk_thread} ${wrk_conn} "
wrk -t ${wrk_thread} -c ${wrk_conn} -d 60 -H "Connection: close" http://${dest_host}/$data
done;

Press the Esc key to exit edit mode, enter :wq, and then press Enter to save and exit the file.

Step 4: Run the test script on the Client and obtain results

Obtain the private IP address of the Server, and then execute the following command on the Client to run the test.

bash wrk_run.sh <serverIP> <threadNums> <connNums>

serverIP (required): Indicates the private IP address of the Server. To obtain the private IP address, go to the Instances page in the ECS console, and then view it in the "Primary IPv4 address" column.

threadNums (optional): Indicates the number of threads to be created. Default value: 128.

connNums (optional): Indicates the number of connections to be created. Default value: 1,024.

The test results are as follows. Pay attention to the Requests/sec metric, because it indicates the average number of requests completed per second. The higher this value, the better the Nginx performance.

Detailed result descriptions are as follows.

Tue Apr 25 02:50:05 PM CST 2023 BENCHMARK STATR to stress nginx_long_ops[hello] with wrk $serverIP 128 1024
Running 1m test @ http://$serverIP/hello
128 threads and 1024 connections (Total 128 threads, 1024 connections)
Thread Stats Avg Stdev Max +/- Stdev
(Mean) (Standard Deviation) (Maximum Value) (Proportion within Standard Deviation)
Latency 1.03ms 182.06us 19.56ms 95.21%
(Latency)
Req/Sec 7.83k 550.46 37.53k 99.67%
(Requests per Second)
59976763 requests in 1.00m, 14.74GB read (Total runtime 1 minute, completed 59,976,763 requests, read 14.74GB of data)
Requests/sec: 997958.48
(Average requests completed per second)
Transfer/sec: 251.21MB
(Average bytes transferred per second)

Optimal configuration

Multiple factors can impact Nginx performance, such as parameter configurations of the Nginx service, kernel parameters, and network card queues. To optimize your ECS instance configurations for optimal Nginx performance, refer to the instructions below.

Disable napi_tx

Disabling napi_tx can reduce the number of hardware and software interrupts when the network interface transmits packets. As a result, Nginx performance can be improved.

This feature is enabled by default only when the Linux kernel version is ≥ 5.3. You can execute the uname -r command to check the kernel version, and then disable napi_tx as follows.

Ubuntu

# Disable napi_tx:
rmmod virtio_net && modprobe virtio_net napi_tx=0 && systemctl restart systemd-networkd.service
# Enable napi_tx:
rmmod virtio_net && modprobe virtio_net napi_tx=1 && systemctl restart systemd-networkd.service

Debian

# Disable napi_tx:
rmmod virtio_net && modprobe virtio_net napi_tx=0 && systemctl restart networking.service
# Enable napi_tx:
rmmod virtio_net && modprobe virtio_net napi_tx=1 && systemctl restart networking.service

CentOS

# Disable napi_tx:
rmmod virtio_net && modprobe virtio_net napi_tx=0 && systemctl restart nmcli c reload
# Enable napi_tx:
rmmod virtio_net && modprobe virtio_net napi_tx=1 && systemctl restart nmcli c reload

Execute the cat /sys/module/virtio_net/parameters/napi_tx command. If N is returned, napi_tx is disabled successfully.

Mitigate contention issues

For large-specification instances, contention issues may occur due to the increased number of CPU cores and thereby impact Nginx performance. These mainly include contention for HTML files and socket contention. Below are solutions for these two types of contention to improve Nginx performance.

Solution 1: Modify the number of Nginx worker processes and enable reuseport

After deploying the Nginx service on the Server, you can enable the Nginx reuseport option to allow multiple sockets to listen on the same port to mitigate contention. Or, specify an appropriate number of Nginx worker processes when HTML files are being accessed to improve Nginx performance. The optimal number of Nginx worker processes varies by instance specification, with details shown below.

Instance type	Optimal number of Nginx worker processes for long connections	Optimal number of Nginx worker processes for short connections
ecs.g3i.2xlarge (8 vCPU)	8	8
ecs.g3i.8xlarge (32 vCPU)	12	6
ecs.g3i.16xlarge (64 vCPU)	13	13

To modify the number of Nginx worker processes and enable reuseport, follow these steps:

Execute the following command to open the /etc/nginx/nginx.conf configuration file.

vim /etc/nginx/nginx.conf

Press i to enter the edit mode.

Modify the number of Nginx worker processes and enable reuseport.

Modify the value of worker_processes based on the specification.

Add the following content to the end of the configuration file to enable reuseport.

http{
server{
listen 80 reuseport;
listen [::]:80 reuseport;
}
}

Press the Esc key to exit edit mode, enter :wq, and then press Enter to save and exit the file.

Execute the following command to restart the Nginx service and apply the configuration.

systemctl restart nginx.service

Solution 2: Install irqbalance-ng

You can follow the guide in network interface interrupt binding to install irqbalance-ng, which automatically configures interrupts for physical network interfaces. This enables interrupt load balancing and thereby enhances network performance.

Disable sniffer

Execute the following command to check whether sniffer is enabled.

cat /proc/net/ptype

If the following output is displayed, the sniffer is not enabled. Then, skip this step.

If the following output is displayed, sniffer is enabled. Proceed to the next step.

Execute the following command to check if the dhclient process is present.

ps aux | grep dhclient

The output is displayed below, which indicates that the dhclient process is present. Execute the following command to terminate the dhclient process and thus disable the sniffer.

kill -9 <process_id>

Last updated: 2026.10.14 17:31:17