当前位置:网站首页>How much faster is a server equipped with a SSD than a mechanical hard disk
How much faster is a server equipped with a SSD than a mechanical hard disk
2020-11-08 16:17:00 【Zhang Yanfei Allen】
I believe you all know that SSD (SSD) The advantage of this is that it is faster than the traditional mechanical hard disk (HDD) Be quick , So now more and more solid-state drives appear in online servers . But as a development that cares about performance numbers , I want to be more precise about carrying SSD The server is in IO The performance is better than that of carrying HDD How much faster , The order IO How much faster in the case of , Random IO How much faster under the circumstances ? Finally, I took time to do a performance test recently .
Test environment
I found two servers , The disks are all Raid array , They are carrying SSD and HDD, More detailed configuration is as follows :
- HDD Real machine :
- Raid The card model is PERC H730 Mini,1GB RAM cache
- 4 block 500GB SSD Composed of raid5, Available capacity 1.3T
- SSD Real machine :
- Raid The card model is the same as above
- 7 block 300G HDD The hard disk makes up Raid5, Available capacity 1.6T
Okay , The contestants are in position , Then select the pressure measuring tool fio. Designated ratio ( Pressure ) " ( measuring ) gauge ( ginseng ) be ( Count ) as follows :
- IO engine : In order to put enough pressure on the hard disk , Choose asynchronous mode libaio
- PageCache: In order to get closer to the performance of the disk itself , Turn on Direct Parameters bypass Linux System memory cache
- Pressure test object : On the safe side , No naked device used , The files used
- file size :100GB, Because of my Raid The card cache is 1G, The guarantee is much bigger than the cache
- Scheduling strategy : Choose common noop
The contrast test officially began .
Sequential read PK
Let's start with a test under sequential access , Because in this environment, whether it's a mechanical hard disk , Or a solid-state drive , Performance will be optimal .IO Size from 512K Until 1M.
Let's take a look at the mechanical hard disk :
Under SSD :
The data looks like , In order IO Under the circumstances , Actually, it carries HDD Server for IO Performance is not being carried ssd It's falling too much . This is because
- The order IO Under the circumstances ,RAID The card has a high hit rate , Especially when it's set up RAID Prefetch
- In its own order IO It's also the best state of mechanical disk work
therefore , Modern in the field of servers ,SSD It's just that there are more and more applications , But it doesn't replace HDD. One of the reasons is because of the order of the mechanical hard disks IO It's not slow , Plus Raid Blessing , So there's a big market share . After comparing the order IO, Let's compare another application scenario , Random IO.
Random reads PK
Remember the beginning fio In the pressure measurement parameters, we put Page Cache It's disabled , Plus the size of the test file is 100GB, This is much more than Raid The cache size of the card . So random based on this premise IO, All levels of cache are basically out of use ,IO Performance speed depends entirely on the hard disk itself .
stay 《 Mechanical hard disk random IO Slower than you think 》 in , We look at the random of mechanical hard disks IO It's really bad performance .
Let's look at carrying SSD Server for IO performance :
By contrast , Mechanical hard disk random IO It's not that slow , We choose the most recognized result of the above 4K IO size Look at .
- From a bandwidth perspective , Only less than 1M The throughput of ,
- From the perspective of delay , stay 4.4ms about
Let's look at the SSD 4K IO Size Next performance . And under the same conditions SSD Can have 32M.
- From a bandwidth perspective , There are still 32M, Than HDD It can be dozens of times faster
- From the perspective of delay , about 120us about . Than HDD, Namely 4400:120, It's also dozens of times higher
summary
In order IO Under the circumstances , Because there is Raid The blessing of card cache , be based on HDD Composed of Raid The array server is in IO The performance is not too bad . But in a random read scenario , Whether bandwidth or latency metrics ,SSD It's better than HDD More than ten times faster . So in the server application SSD, The main solution is random IO The problem of . So if you have too many Page Cache,Raid Card cache can't hold random IO request , Then please put HDD Replace it with SSD Well .
Development of hard disk album of internal training :
- 1. Disk opening : Take off the hard coat of the mechanical hard disk !
- 2. Disk partitioning also implies technical skills
- 3. How can we solve the problem that mechanical hard disks are slow and easy to break down ?
- 4. Disassemble the SSD structure
- 5. How much disk space does a new empty file take ?
- 6. Only 1 How much disk space does a byte file actually take up
- 7. When there are too many documents ls Why is the command stuck ?
- 8. Understand the principle of formatting
- 9.read How much disk does a byte of file actually take place on IO?
- 10.write When to write to disk after one byte of file IO?
- 11. Mechanical hard disk random IO Slower than you think
- 12. How much faster is a server equipped with a SSD than a mechanical hard disk ?
My official account is 「 Develop internal skill and practice 」, I'm not just talking about technical theory here , It's not just about practical experience . It's about combining theory with practice , Deepen the understanding of theory with practice 、 Use theory to improve your technical practice ability . Welcome to my official account , Please also share with your friends ~~~
版权声明
本文为[Zhang Yanfei Allen]所创,转载请带上原文链接,感谢
边栏推荐
猜你喜欢
Gopherchina 2020 Conference
Travel notes of Suzhou
It's just right. It's the ideal state
LiteOS-消息队列
机械硬盘随机IO慢的超乎你的想象
read文件一个字节实际会发生多大的磁盘IO?
[open source]. Net uses ORM to access Huawei gaussdb database
We made a medical version of the MNIST dataset, and found that the common automl algorithm is not so easy to use
VIM configuration tutorial + source code
刚刚好,才是最理想的状态
随机推荐
python开发qt程序读取图片的简单流程
2035我们将建成这样的国家
[open source]. Net uses ORM to access Huawei gaussdb database
第五章编程题
刚刚好,才是最理想的状态
阿里云加速增长,进一步巩固领先优势
Learn to record and analyze
Improvement of maintenance mode of laravel8 update
AI weekly: employees are allowed to voluntarily reduce salary; company response: employees are happy and satisfied; tiger tooth HR takes employees out of the company; Sweden forbids Huawei ZTE 5g equi
Chapter 5 programming
. net large data concurrency solution
Station B STM32 video learning
Solution to the problem of offline connection between ADB and mobile phone
大龄程序员没有出路吗?
阿里云视频云技术专家 LVS 演讲全文:《“云端一体”的智能媒体生产制作演进之路》
Xiaoqingtai officially set foot on the third day of no return
机械硬盘随机IO慢的超乎你的想象
Golang ICMP协议探测存活主机
Arduino IDE搭建ESP8266开发环境,文件下载过慢解决方法 | ESP-01制作WiFi开关教程,改造宿舍灯
The birth of a new integrated memory and computing chip is conducive to the application of artificial intelligence~