Cloud vendor virtual machine performance test

Test tools and main indicators

Classification	test tools	Test indicators
CPU	UnixBench	single core score
	UnixBench	Multi-core running score
	SuperPI	Calculation time
Memory	Stream	bandwidth
Memory	MLC	time delay
network	iPerfs	bps、pps
	Pktgen
	Netperf
cloud disk	FIO	IOPS
	FIO	I/O latency
	dd	bandwidth

test environment

Preparation

Register a Volcano Engine account and complete the enterprise real-name authentication . You can log in to the Volcano Engine console , click "Account Management > Real-name Verification" on the right side of the top navigation bar, and go to the real-name verification page to check whether the real-name verification has been completed.
Activate the following services:
- Cloud server ECS
- Private network VPC

Environmental preparation

To create a VPC , it is recommended to select China North 2 (Beijing) as the region .
Create two ECS cloud servers of the same specification under the same VPC, select one of them to bind the public network IP and mount a data cloud disk (not less than 2T is recommended).

illustrate

The test environment and methods given in this article are for reference only, and do not represent the actual optimal performance of the instance.

Software version

Cloud server operating system: This article takes CentOS 8.3 as an example.

Test Methods

CPU performance test

UnixBench test CPU single-core/multi-core performance

Test instruction

UnixBench is an open source tool for testing the basic performance of UNIX systems, and a relatively general tool for testing the performance of cloud servers. UnixBench will perform a series of tests to test the performance of various aspects of the system, and then compare each test result with a benchmark value to obtain an index value. The index values of all test items are combined to form a test score value.

test steps

Execute the following command to download and install UnixBench.

wget https://github.com/kdlucas/byte-unixbench/archive/v5.1.3.tar.gz
tar -zxvf v5.1.3.tar.gz
cd byte-unixbench-5.1.3/UnixBench
make

Execute the following command to perform a single-core test.
./Run -c 1

Please wait patiently after executing this command, the process may be long, and the final score will be displayed on the screen. The result refers to the Index value, the higher the value, the better the performance.
Execute the following command to perform a multi-core test.
```
# 获取 CPU 的线程数 core
cat /proc/cpuinfo| grep "processor"| wc -l
./Run -c {core}
```
Please wait patiently after executing this command, the process may be long, and the final score will be displayed on the screen. The result refers to the Index value, the higher the value, the better the performance.

SuperPI test calculation time

Test instruction

SuperPI is a computer program that calculates pi with a maximum accuracy of 32 million digits after the decimal point. It is a quick reference benchmark often used for overclocking or comparing different CPU and hardware configurations.

test steps

Execute the following commands to install and test the SuperPI tool.

git clone https://github.com/Fibonacci43/SuperPI.git
cd ./SuperPI
gcc -O -funroll-loops -fomit-frame-pointer pi_fftcs.c fftsg_h.c -lm -o pi_css5
# 通过计算精度为 1<<22=4194304 的 π 的值，测试 CPU 的计算能力
./pi_css5 $((1<<22))

If you are prompted that Git is not installed, you can execute the following command to install it.

# CentOS下使用
yum install git -y

# Ubuntu/Debian下使用
apt-get install git -y

After executing the above command, the calculation time is echoed as shown in the figure below. The smaller the value, the shorter the time and the stronger the calculation ability.

memory performance test

Stream test bandwidth

Test instruction

Stream is one of the industry-recognized comprehensive memory performance measurement tools. Sream supports four operations: Copy, Scale, Add, and Triad.

For more information, please go to Stream official website address and Stream usage instructions .

test steps

Run the following commands to download and install Stream.

wget https://www.cs.virginia.edu/stream/FTP/Code/stream.c
yum -y install gcc

Write test scripts.

gcc -O -fopenmp -mcmodel=medium -DSTREAM_ARRAY_SIZE=200000000 -DNTIMES=100 stream.c -o stream_omp_exe

Parameter Description:

parameter	illustrate
STREAM_ARRAY_SIZE	Specifies the size of the array used by the test. The default is 10000000. Stream will create an array of 3 `STREAM_ARRAY_SIZE` sizes, and the array element type is double by default. The memory used in the above reference script is 3 × STREAM_ARRAY_SIZE × sizeof(double)=4.5GiB, which can also be modified according to actual needs.
NTIMES	Specify the number of times to run the test case, and the final result is the best one. The value defaults to 10 and usually does not need to be modified.

Execute the following command to test the bandwidth.

# 同 numa 内存带宽 
numactl -N 0 -m 0 ./stream_omp_exe 
# 跨 numa 内存带宽 
numactl -N 1 -m 0 ./stream_omp_exe

Parameter Description:

parameter	illustrate
numactl	NUMA policy that controls process and memory allocation. For example, `-N 0 -m 0`it indicates the use of NUMA0 CPU and NUMA0 memory, and `-N 1 -m 0`indicates the use of NUMA1 CPU and NUMA0 memory.

After executing this command, the final result score is displayed as shown in the echo in the figure below. The results refer to the Best Rate value, and the higher the value, the better the performance.

MLC Test Latency

Test instruction

MLC (Memory Latency Checker) is Intel's memory testing tool, which can effectively and conveniently test memory latency and bandwidth. The tool also provides options for a range of fine-grained tests from specific processor cores to cache or memory.

test steps

Run the following commands to download and install MLC.

wget https://software.intel.com/content/dam/develop/external/us/en/protected/mlc_v3.9.tgz
yum install glibc -y
tar -xzvf mlc_v3.9.tgz

Execute the following command to perform a delay test.

cd ./Linux/
./mlc --latency_matrix -e -r

Parameter Description:

parameter	illustrate
--latency_matrix	Print same-NUMA and cross-NUMA memory latency matrix
-e	output test results that do not modify prefetcher settings
-r	Output the delay result obtained by random access

After executing this command, the matrix of local and cross-socket memory latency will be printed on the screen. The lower the value, the lower the latency.

Network Performance Test

See Testing Network Performance Using iPerf3 , Testing Network Performance Using Pktgen , Testing Network Performance Using Netperf .

Cloud Disk Performance Test

See Testing storage performance .

Transferred from:

Cloud Server Performance Benchmark--Cloud Server-Volcano Engine