Linux性能测试和优化实践笔记

前言

Linux性能测试、监控、优化是一个持续的过程，上图为LinuxCon上Brendan D. Gregg分享的Linux benchmarking tools示意图，涵盖面十分广泛。我们可以通过成熟的监控方案如Prometheus，Zabbix来捕获大部分信息，在实际工作中我们会经常关注CPU、Memory、I/O、Network等性能问题，本文增加了倪朋飞在极客时间出品的《Linux性能优化实战》专栏课程学习笔记。

Donald Knuth说”过早优化是万恶之源”（premature optimization is the root of all evil)

---

更新记录

2022年10月18日 - 增加Linux性能优化实战
2022年08月31日 - 增加bench.sh和yabs.sh
2015年03月06日 - 初稿

阅读原文 - https://wsgzao.github.io/post/linux-performance/

扩展阅读

Linux Performance - http://www.brendangregg.com/linuxperf.html

---

CPU

确认CPU型号

cat /proc/cpuinfo |grep "model name"|uniq|cut -f2 -d:
    
Intel(R) Xeon(R) CPU E5-2650 v2 @ 2.60GHz

Super PI

计算时间越短越好

time echo "scale=500;4*a(1)"|bc -l -q
    
3.141592653589793238462643383279502884197169399375105820974944592307\
81640628620899862803482534211706798214808651328230664709384460955058\
22317253594081284811174502841027019385211055596446229489549303819644\
28810975665933446128475648233786783165271201909145648566923460348610\
45432664821339360726024914127372458700660631558817488152092096282925\
40917153643678925903600113305305488204665213841469519415116094330572\
70365759591953092186117381932611793105118548074462379962749567351885\
75272489122793818301194912
    
real    0m0.081s
user    0m0.076s
sys     0m0.000s

Disk

清空缓存

每次做读写测试前建议先清空缓存

sync; echo 3 > /proc/sys/vm/drop_caches

测试读性能

选择测试磁盘，建议做2-3组取平均值

hdparm -t /dev/sda

/dev/sda:
Timing buffered disk reads:  1074 MB in  3.00 seconds = 357.92 MB/sec

测试写入性能

根据业务选择不同的BlockSize大小按需多次测试取平均值

time dd if=/dev/zero of=/tmp/speed bs=1M count=2K conv=fsync;rm /tmp/speed

参考数据

以10,000 rpm 300 GB SAS硬盘为例，机型为IBM x3650 M4，Raid参数如下

1.Read Policy：Ahead （控制器缓存读策略：预读）
2.Write Policy：Write Back with BBU（控制器缓存写策略：有电池备份时回写）
3.IO Policy: Direct（IO策略：直接）
4.Drive Cache：disable （硬盘缓存：禁用）

Raid	Read(MB)	Write(MB)
Raid 1	170	130
Raid 5	350	250
Raid 10	300	215

---

开源的一键测试工具

bench.sh

一键测试脚本bench.sh

1、显示各种系统信息；
2、取自 Speedtest 世界多处的数据中心的测试点，网络测试比较全面；
3、支持 IPv6 下载测速；
4、IO 测试（顺序写入 1GB 数据）三次，并显示其平均值。

再配合 unixbench.sh 脚本测试，即可全面测试 VPS 的性能。
https://teddysun.com/245.html

Description: Auto test I/O & upload & download speed script
Intro: https://teddysun.com/444.html

wget -qO- bench.sh | bash

curl -Lso- bench.sh | bash

yabs.sh

Here’s an attempt to create yet another damn Linux server benchmarking script.

https://github.com/masonr/yet-another-bench-script

curl -sL yabs.sh | bash

wget -qO- yabs.sh | bash

This script has been tested on the following Linux distributions: CentOS 6+, Debian 8+, Fedora 30, and Ubuntu 16.04+. It is designed to not require any external dependencies to be installed nor elevated privileges to run.

Local fio/iperf3 Packages: If the tested system has fio and/or iperf3 already installed, the local package will take precedence over the precompiled binary.

Experimental ARM Compatibility: Initial ARM compatibility has been introduced, however, is not considered entirely stable due to limited testing on distinct ARM devices. Report any errors or issues.

High Bandwidth Usage Notice: By default, this script will perform many iperf network tests, which will try to max out the network port for ~20s per location (10s in each direction). Low-bandwidth servers (such as a NAT VPS) should consider running this script with the -r flag (for reduced iperf locations) or the -i flag (to disable network tests entirely).

VPS常用测试工具

VPS常用测试脚本
https://zhuanlan.zhihu.com/p/117547388

流媒体检测脚本
https://github.com/sjlleo/netflix-verify
https://github.com/lmc999/RegionRestrictionCheck

Linux性能优化实战

Linux 性能优化实战
倪朋飞 资深 Linux 专家，Kubernetes 项目维护者
https://time.geekbang.org/column/intro/100020901

Linux性能优化实战笔记

Linux performance optimization
https://feiyang233.club/post/linux/