Performance comparison of ext4, NTFS, XFS, Btrfs, ZFS, f2fs and ReiserFS

ext4、ntfs、xfs、btrfs、zfs、f2fs and reiserFS Performance test comparison

Original address ：CSDN Blog

---

---

1. Preface

Current file systems are diverse , From the well-known ext4、ntfs and xfs, To btrfs、zfs Wait for the niche file system , It can be said that there are a wide range , It is difficult for people to choose a file system suitable for their business .

This article summarizes the foreign netizens in OpenBenchmark Results on the website , And give the performance and comparison of major file systems in different situations .

The file systems involved at present include ：

• btrfs
• xfs
• ext4
• f2fs
• reiserFS
• ntfs
• zfs

Writing is not easy , Still hope One key, three links Pay attention ！

---

2. Linux 4.4 Benchmark

The benchmark data comes from Linux 4.4 File-Systems.

2.1 Test environment

name	To configure
processor	Intel Xeon E5-2687W v3 @ 3.5GHz(20 Cores)
a main board	MSI X99S SLI PLUS(MS-7885) v1.0
Chipset	Intel Xeon E7 v3
Memory	16GB
Hard disk	PNY CS1211 120GB + 80GB Intel SSDSCKGW08
operating system	Ubuntu 16.04
Kernel version	4.4.0-040400-generic(x86_64)
GCC	GCC 5.3.1 20160122

Various file system mount options ：

file system	Mount options
btrfs	relatime,rw,space_cache,ssd,subvol=/,subvolid=5
xfs	attr2,inode64,noquota,relatime,rw
ext4	data=ordered,relatime,rw
f2fs	acl,active_logs=6,background_gc=on,extent_cache,inline_data,relatime,rw,user_xattr
reiserFS	relatime,rw
ntfs	allow_other,blksize=4096,group_id=0,relatime,rw,user_id=0
zfs	defaults

2.2 Test means

For the file system benchmark, It is divided into the following test methods ：

• aio-stress： Use SuSE Asynchronous development I/O frame , Test asynchrony I/O Under high pressure Random write Ability .
• sqlite： measurement SQLite The database has a fixed number of Insert Operation of the write in when , This is a The smaller the value, the better Test of .
• flexible IO tester： Write directly to disk without cache , Test the hard disk pair 4KB The size of the block Random, speaking, reading and writing And Sequential reading and writing Ability .
• fs-mark： Create many 1MB File to test the file system write in Ability .
• dbench： Simulate multi-user read-write pressure benchmark Tools .
• compilebench： Through the compilation and update of the kernel tree , Simulate and test the performance after the hard disk is almost full and the folder is stored for a long time .

2.3 test result

The test results are summarized as follows , Among them ^ In the same test Optimal data , mark * by Worst data . The data unit is MB/s Or seconds , And only the integer part is reserved .：

test	btrfs	xfs	ext4	f2fs	reiserFS	ntfs	zfs
aio-stress	^2920	2385	2424	1606	1606	11*	146
sqlite ( The smaller the better. )	755*	379	384	471	384	^294	501
fio: Random reads	104	109	110	104*	112	^215
fio: Random write	84	80	79	95	75*	^98
fio: Sequential read	113*	185	185	184	185	^219
fio: Write in sequence	81*	93	87	84	96	^101
fs-mark: 1K file	35	35	35	^60	31*	35	37
fs-mark: 5K file , 4 Threads	68	71	^112	76	79	41*	59
fs-mark: 5K file , 32 subdirectories	35	35	36	^60	31*	34	36
dbench: 6 Customer	53	^62	61	54	47	29*	60
dbench: 12 Customer	105	^118	107	96	74	30*	117
compilebench: Create a kernel tree	160	206	314	^339	115	44*	145

The chart of each test is as follows ：

2.4 Comparison of major file systems

1. High concurrent read and write ：btrfs The optimal ,xfs And ext4 Slightly better ,zfs And ntfs The worst .
2. Random, speaking, reading and writing ：ntfs The optimal , The rest are not much different .
3. Sequential reading and writing ：ntfs The optimal ,btrfs The worst , The rest are not much different
4. Sequential multi file write ：f2fs The optimal , The rest are not much different .
5. Concurrent multi file writing ：ext4 The optimal ,ntfs The worst ,zfs Weak , The rest are not much different .
6. Sequential multiple files are written to different subdirectories ：f2fs The optimal , The rest are not much different .
7. Multi customer experience ：xfs The optimal ,ntfs The worst ,reiserFS Weak , The rest are excellent .

Through this benchmark, You can find Each file system has characteristics , And each has its own advantages and disadvantages ：

• btrfs Still in development , Its COW (Copy On Write) The mechanism makes it perform poorly in inserting into the database .
• xfs and ext4 Excellent comprehensive quality , It won't be too dazzling in certain abilities , But for users Comprehensive experience is the best .
• f2fs Excellent performance in multi document situations .
• reiserFS Is about to be reiser4 File system replaces , The comprehensive performance is not as good as other systems .
• ntfs In the serial operating mode , Its Both random and sequential reading and writing abilities are extremely high ; conversely , Its presence Poor performance under high concurrent and high pressure environment .

notes ：fio Missing in the test zfs related data , It will be re analyzed in subsequent tests .

---

3. ZFS Ubuntu 16.04 test

because zfs Relevant data are in 2 The test of chapter is not perfect , Therefore, this chapter adds zfs Relevant test contents .

The data in this chapter refer to ZFS Ubuntu 16.04 Testing.

3.1 Test environment

name	To configure
processor	Intel Xeon E5-2687W v3 @ 3.5GHz(20 Cores)
a main board	MSI X99S SLI PLUS(MS-7885) v1.0
Chipset	Intel Xeon E7 v3
Memory	16GB
Hard disk	PNY CS1211 120GB + 80GB Intel SSDSCKGW08
operating system	Ubuntu 16.04
Kernel version	4.4.0-21-generic (x86_64)
GCC	GCC 5.3.1 20160122

Various file system mount options ：

file system	Mount options
zfs	defaults
ext4	data=ordered,relatime,rw
btrfs	relatime,rw,space_cache,ssd,subvol=/,subvolid=5
xfs	attr2,inode64,noquota,relatime,rw

notes ： This environment and testing 2 Basically the same in , You can jointly view .

3.2 Test means

For the file system benchmark, It is divided into the following test methods ：

• fs-mark： Create many 1MB File to test the file system write in Ability .
• blogbench： Simulate the pressure scenario of a blog website , test Random, speaking, reading and writing The ability of .
• dbench： Simulate multi-user read-write pressure benchmark Tools .
• iozone： Launch many different types of I/O request , Test the performance of the file system .
• compilebench： Through the compilation and update of the kernel tree , Simulate and test the performance after the hard disk is almost full and the folder is stored for a long time .、

3.3 test result

And 2.3 The test style in is basically the same , The unit used is MB/s Or other special units ：

test	xfs	ext4	btrfs	xfs
fs-mark: 1K file	^37	35	35	34*
fs-mark: 5K file , 4 Threads	59*	^112	67	69
fs-mark: 5K file , 32 subdirectories	^36	35	35*	35
blogbench: Read	358K*	563K	2719K	^3688K
blogbench: write in	906*	^13086	7602	5382
dbench: 6 Customer	59	^61	52*	60
iozone: 4Kb, 8GB Read	1131*	^6272	6225	6138
iozone: 4Kb, 8GB write in	201*	464	^478	464
compilebench	178*	^372	191	252

3.4 zfs Performance analysis

Compared with other commonly used file systems ,zfs You can't take too many advantages in performance , By contraries ,zfs It is inferior to other file systems in all aspects .

But that doesn't mean zfs Good for nothing ,zfs The usage scenario of is mainly located in the multi hard disk system , Its most distinctive function is It provides file system level raid Array capability , Therefore, there are still many applicable scenarios .

---

4. About performance testing

Performance testing does not represent the actual experience of file systems , Before readers read this article , This point needs to be clarified .

Many file systems are not necessarily designed for performance , They all have their own scenarios , Each file system has its own unique design intention . for example ：

• btrfs and zfs All have COW(Copy On Write) Mechanism .COW The mechanism makes them have excellent performance when modifying , But it also brings a waste of disk space , And increases the consumption of reading .
• btrfs and zfs Have their own raid function . When data availability is required , Compared with other file systems , They can provide out of the box raid Array capability .
• btrfs All data are provided with checksum Mechanism , Ensure file integrity .
• xfs It's dynamic inode Allocate capacity , Suitable for large file allocation .
• … …

Adjust measures to local conditions , Just used , Is the way of programmers .

---

5. summary

The author summarizes OpenBenchmark Some of the data on , And integrated the performance comparison of file system .

There are not many words , Easy to use is OK. . The author has just returned CSDN It didn't take long , Feeling CSDN In writing , Filled with thick employment 、 Exam atmosphere . For bloggers like me who just want to watch cutting-edge technologies and novel reports , Not very friendly , And I have a hunch that this article will also throw stones into the sea , Can't lift a storm .

It's ok , Only those who are useful can take it , May the spirit of open source survive ！

If you find this article useful , Still hope One key, three links , I must be very grateful ！