Skip to content

Instantly share code, notes, and snippets.

Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save kokizzu/1b6bd230a9e9677d5ee89cc80ce7557a to your computer and use it in GitHub Desktop.
Save kokizzu/1b6bd230a9e9677d5ee89cc80ce7557a to your computer and use it in GitHub Desktop.
which file sytem to use for daily work? should we turn on btrfs compression?
#btrfs benchmark for daily used desktop OS
*.backup
/data

Introduction

BTRFS offers lots of advantages such as snapshot, self-healing from bit rotten, compression, etc. There are lots of benchmarks and reviews about its benefits and weakness. However, almost all of them focus on very low level performance such as sequential/random write/read, etc. Or they emphasize special use case such as SQL, game loading or system boot up time. But it is difficult to interpret them. What is the impact on our daily work as developers? Moreover all of them only compare different system on same hardware and assume the results can be extrapolated in general regardless the hardwares. Unfortunately, from my user experience, the file systems behave completely different on different hardwares.

In this article I would like to Compare performance of current popular file systems: BTRFS, XFS and EXT4 based on normal daily use cases for developers: no random writing on single file, lots of small files in build dir. What is the best choice? Do the choices vary when using different hardwares: SATA SSD, NVME SSD, mechanic HDD, weaker CPUs? Does LUKS impose unacceptable performance drop?

The BTRFS also come with important features such as compression. However, compression of file system in most cases are nonsense nowadays. It won't save any space or increase IO performance. Because normal files in PC/laptop are all compressed already: (image, pdf, media, xls/doc, hdf, ccache, read-only database ...). I think only super large build directory benifits from file system compression. The git repo itself actually has compression already. Thus if the .git is way larger than source tree, it won't make too much sense to compress the entire source dir. Fortunately, the btrfs should be smart enough to determine which files are worth compressing.

The result is rather surprising: the performance varied a lot depends on what kind of hardware I am using and what kind of data I have; the worst choices can be any of those. However, the best choice seems remain the same: btrfs with mild compression.

methods

  • write test: cp -a <src> <dest> && sync
  • before write test I delete the old items in <dest> and fstrim the mount point.
  • tar read test: tar -c <dest>/<data> | pv -f /dev/null
  • cp read test: cp -a <dest> /tmp/<test root>
  • before each read test we remount the partition to drop the system cache

1 https://blog.cloudflare.com/speeding-up-linux-disk-encryption/ [2] https://www.reddit.com/r/Fedora/comments/rzvhyg/default_luks_encryption_settings_on_fedora_can_be/ [3] https://www.reddit.com/r/Fedora/comments/rxeyhd/fastest_filesystem_with_encryption_for_development/

Results

compression ratio

For simple benchmark, I used several ffmpeg and libboost build dirs as example: ffmpeg

LZO
Processed 1541 files, 20427 regular extents (20427 refs), 542 inline.
Type       Perc     Disk Usage   Uncompressed Referenced  
TOTAL       47%      1.2G         2.6G         2.6G       
none       100%      306M         306M         306M       
lzo         41%      970M         2.3G         2.3G       
ZLIB
Processed 1541 files, 20399 regular extents (20399 refs), 632 inline.
Type       Perc     Disk Usage   Uncompressed Referenced  
TOTAL       36%      964M         2.6G         2.6G       
none       100%      294M         294M         294M       
zlib        28%      670M         2.3G         2.3G       
ZSTD:1
Processed 1541 files, 20399 regular extents (20399 refs), 632 inline.
Type       Perc     Disk Usage   Uncompressed Referenced  
TOTAL       34%      917M         2.6G         2.6G       
none       100%      294M         294M         294M       
zstd        26%      622M         2.3G         2.3G       
ZSTD:3
Processed 1541 files, 20399 regular extents (20399 refs), 632 inline.
Type       Perc     Disk Usage   Uncompressed Referenced  
TOTAL       34%      917M         2.6G         2.6G       
none       100%      294M         294M         294M       
zstd        26%      622M         2.3G         2.3G  

libboost dir (149k/2GiB) contains lots of small files. The apparent size of du is 1.6GB, but the actual disk size is 2.0GB, implying that there are lots of file smaller than 4kiB:

LZO
Processed 148537 files, 60482 regular extents (60482 refs), 92581 inline.
Type       Perc     Disk Usage   Uncompressed Referenced  
TOTAL       37%      647M         1.6G         1.6G       
none       100%      4.3M         4.3M         4.3M       
lzo         37%      643M         1.6G         1.6G       

ZLIB
Processed 148537 files, 59862 regular extents (59862 refs), 93201 inline.
Type       Perc     Disk Usage   Uncompressed Referenced  
TOTAL       24%      417M         1.6G         1.6G       
none       100%      228K         228K         228K       
zlib        24%      416M         1.6G         1.6G       

ZSTD:1
Processed 148537 files, 59862 regular extents (59862 refs), 93201 inline.
Type       Perc     Disk Usage   Uncompressed Referenced  
TOTAL       22%      392M         1.6G         1.6G       
none       100%      228K         228K         228K       
zstd        22%      392M         1.6G         1.6G       

ZSTD:3
Processed 148537 files, 59862 regular extents (59862 refs), 93201 inline.
Type       Perc     Disk Usage   Uncompressed Referenced  
TOTAL       22%      392M         1.6G         1.6G       
none       100%      228K         228K         228K       
zstd        22%      392M         1.6G         1.6G

So ZSTD compress the data pretty well. LZO provide less compression but still decent ~50%. I also noticed the ZLIB impose huge pressure on CPU and very slow. So in the following tests I ignored it.

preliminary tests

I also tested the speed or the write/read via simple cp -a <source> <dest> on a Intel Core i7-9750H machine.

write test to file system on a same usb SSD (files are already read into memory):

dest time (sec)
xfs 1.1
btrfs 0.97
btrfs zstd:1 0.94

We can see the speed is almost the same. This indicates the penality on write performance is quite huge when zstd is on. Because it actually only write 1/3 of the data.

read test from above file system to a ram disk, system cache is dropped before each copy command.

source time (sec) peak speed (M/s)
xfs 7.7 366.1
btrfs 6.7 379.0
btrfs zstd:1 2.9 338.9

peak speed is obtained via iostat. With zstd:1 the I/O is slower but not much comparing to the write operation. However, the big compression ratio compensate this and yield the fastest read speed. It is 130% faster than non-compressed condition, since the size is 1/3 of the original.

I did not test other combination or zlib, since the zlib is well-known slower than zstd:1

However, this test has a flaw that I do not sync after cp, thus the data may not finished the write-operation to the disk at all after cp command finished. In following more thourough tests, I use cp xxx && sync as write speed test.

results on usb3 SATA SSD

/dev/sdd:480103981056B:scsi:512:512:gpt:INTEL SS DSC2BW480A4:;
1:17408B:1073741823B:1073724416B:free;
1:1073741824B:27917287423B:26843545600B:ext4:usb ssd test ext4:;
3:27917287424B:54760833023B:26843545600B:btrfs:usb ssd test btrfs:;
5:54760833024B:81604378623B:26843545600B:xfs:usb ssd test xfs:;

This disk has about 300MiB/s sequencial write speed and 500MiB/s sequencial read speed.

Writing 1.4k files in size of 2.7GiB on all file systems gave similar write speed as writing single big file. the non-compressed btrfs has a bit less write and read performance than xfs and ext4. Compressed btrfs give way better performance. zstd:3 is the best with 55% more write speed and 133% more read speed. It is followed by zstd:1 with 35%/133% increase. The LZO has less improvement but still give 43% more write speed and 79% more read speed.

index file system disk usage write time (s) write MiB/s write rate tar to file read MiB/s tar read rate cp read MiB/s cp read rate
0 btrfs.none 0.000147705 9.40651 283.958 0.965568 365.276 1.07789 344.891 0.958941
5 xfs.0 0.112771 9.38198 284.701 0.968093 341.466 1.00763 363.048 1.00942
2 ext4.0 1.40477e-06 9.2048 290.181 0.986728 339.391 1.00151 360.402 1.00207
3 xfs.1 0.00823492 9.1085 293.249 0.997159 338.185 0.997949 363.603 1.01097
6 ext4.1 0.106864 8.96366 297.987 1.01327 338.369 0.998491 358.914 0.997932
1 btrfs.zstd-1 0.104891 6.71203 397.95 1.35319 253.149 0.747018 838.609 2.33168
4 btrfs.lzo 0.141149 6.34291 421.109 1.43193 267.076 0.788114 643.236 1.78846
7 btrfs.zstd-3 0.191489 5.83131 458.054 1.55756 250.175 0.738242 853.062 2.37187

Interestingly, tar read test yield completely different scheme: none compressed btrfs give the best reasult, all compressed btrfs give 30% drop in speed.

Writing 149k/2GB files on all file systems is way slower: ~30% of normal writing speed and ~10% of normal reading speed. XFS seems perform worst when many small files invovled. btrfs in general proivdes higher performance (> 18% write and 21% read improvement over ext4) in this case. Unlike the 1.4k-file test, all 3 compressed methods can yielded best write performance providing 35% more speed over ext4, from time to time. It seems like it depends on the test order, the first compressed btrfs yield the best result. The non-compressed one is also quite similar to the best one.

The read speeds are different though. All 3 compressed ones have almost same reading speed with about 60% increase. But the non-compressed one is a bit slower with only 22% improvement over ext4. And in this case the tar read speed trend is the same as cp read speeds. Though tar has about 20% more speed when using compressed btrfs comparing to the cp command.

index file system disk usage write time (s) write MiB/s write rate tar to file read MiB/s tar read rate cp read MiB/s cp read rate
4 xfs.0 0.0879229 13.7472 118.552 0.914611 47.178 0.962071 47.7707 0.978131
1 xfs.1 0.00823614 13.5625 120.166 0.927067 47.6865 0.97244 47.3504 0.969526
5 ext4.0 0.079824 12.6361 128.977 0.995037 49.383 1.00704 49.2813 1.00906
2 ext4.1 1.40477e-06 12.5113 130.263 1.00496 48.693 0.992965 48.3962 0.990938
6 btrfs.lzo 0.106342 11.269 144.622 1.11574 88.2133 1.79888 77.5018 1.58689
7 btrfs.zstd-3 0.143783 10.3497 157.469 1.21485 85.4793 1.74312 80.4497 1.64725
3 btrfs.none 0.0261482 10.1429 160.68 1.23962 64.454 1.31437 59.4803 1.21789
0 btrfs.zstd-1 0.000147705 9.30068 175.23 1.35187 92.5516 1.88734 79.568 1.6292
index file system disk usage write time (s) write MiB/s write rate tar to file read MiB/s tar read rate cp read MiB/s cp read rate
1 xfs.1 0.00823492 14.5515 111.999 0.873386 48.1493 0.992781 49.1477 1.02168
5 xfs.0 0.0879229 13.7274 118.723 0.925814 46.7736 0.964414 48.3584 1.00527
2 ext4.1 0.079824 12.9215 126.128 0.98356 48.5128 1.00027 48.6915 1.0122
0 ext4.0 1.40477e-06 12.5035 130.344 1.01644 48.4862 0.999726 47.518 0.987803
6 btrfs.zstd-1 0.117458 12.0904 134.798 1.05117 92.4985 1.90721 79.8751 1.66044
7 btrfs.zstd-3 0.143814 10.358 157.342 1.22697 85.3323 1.75945 80.1055 1.66523
3 btrfs.none 0.000147705 9.4264 172.893 1.34824 65.216 1.34467 58.5836 1.21784
4 btrfs.lzo 0.0801518 9.07413 179.605 1.40058 86.4693 1.78289 76.8859 1.5983

results on faster SSD

I then test it on an almost fresh nvme SSD:

/dev/nvme0n1:512110190592B:nvme:512:512:gpt:INTEL SSDPEKNW512G8:;
6:296470183936B:350157275135B:53687091200B:xfs:linux-nvme-data:;
7:350157275136B:377000820735B:26843545600B:btrfs:btrfs test:;
8:377000820736B:403844366335B:26843545600B:ext4:ext4test:;

For a 2.7GiB ffmpeg build dir containing 1.5k files, the speeds are all > 500MB/s which is close to the single large file copy speed (~600-700MB/s). I test it several times with random shuffled test orders. The typical results are showing in following table. We can see lzo always has the fastest write speed, though the compression ratio is only 47%. The ZSTD:1 has much higher ratio of 34% however it seems to pay a big penality. Unlike lzo/zstd, ZLIB shows very high CPU usage and very slow in the prelimiary tests. So I did not include it in the repeated tests. The ext4 and xfs speeds yield from this test are kind of unstable, some times to drop to 100-200MB/s. But in most cases they are around 500-700MB/s

file system disk usage write time (s) write MiB/s tar to file read MiB/s cp read MiB/s
btrfs.zstd-1 0.401214 3.14575 849.1 489.592 1305.49
btrfs.lzo 0.437475 1.91317 1396.14 589.98 1280.35
btrfs.none 0.487829 3.29185 811.416 1216.99 1153.71
ext4.0 0.40917 4.48444 595.628 1093.44 1213.11
ext4.1 0.516032 4.4532 599.806 1100.46 1186.14
xfs.0 0.0076032 3.88385 687.734 1190.83 1292.62
xfs.1 0.0598712 5.31797 502.27 1182.8 1271.12

For a 2.0GiB boost build dir containing 149k files, the I/O speeds are far slower around 200MB/s in all cases. The btrfs.lzo and non-compressed btrfs are only slightly faster than ext4 and xfs during writing. But the zstd:1 is way slower than the rest, only 60-70% speed of non-compressed btrfs. The read speed of all btrfs tests are 20% faster than ext4 and xfs. This may indicate btrfs handle small files better. The compression ratio in this case does not affect the I/O much because the bottle neck is the ammount of files instead of size, though LZO and ZSTD:1 provide 37% and 22% ratios respectively.

file system disk usage write time (s) write MiB/s tar to file read MiB/s cp read MiB/s
btrfs.lzo 0.401216 8.13217 200.408 163.573 124.486
btrfs.none 0.437983 7.60417 214.324 146.083 120.017
btrfs.zstd-1 0.518031 13.0737 124.659 153.742 121.545
ext4.0 0.40917 8.51435 191.413 96.7245 107.22
ext4.1 0.488992 8.64948 188.422 96.3132 107.352
xfs.0 0.0076032 8.57081 190.152 110.243 102.137
xfs.1 0.0474472 8.57814 189.989 110.808 100.598

An interesting observation is that tar read speed and the cp read speed are all very different in both 1.5k/2.7GiB and 148k/2GiB tests. However, the effects are in a different direction. In 1.5k/2.7GiB case the LZO/ZSTD:1 show ~50% performance drop in tar-read comparing to cp tests of themselves and the tar tests of other file systems. This indicates the file reading methods of tar and cp have some fundamental difference and worth further study.

In 148k/2GiB case the compressed file systems do not show any performance drop in tar tests. However, tar tests of btrfs are ~25% faster than cp tests. The possible reason can be that large number of file creation operations in RAM disk is still costy.

On mechanic HDD

/dev/sdc:5000981077504B:scsi:512:4096:gpt:Seagate One Touch HDD:;
1:17408B:107374182399B:107374164992B:free;
1:107374182400B:134217727999B:26843545600B:ext4:speed test ext4:;
2:134217728000B:161061273599B:26843545600B:xfs:speed test xfs:;
3:161061273600B:187904819199B:26843545600B:btrfs:speed test btrfs:;
./fs-user-benchmark.py --no-fstrim --source ~/.cget/cache/builds/ffmpeg -t '{"root_path":"/media/dracula/hddxfstest/test", "postfix":"0"}' -t '{"root_path":"/media/dracula/hddext4test/test", "postfix":"0"}' -t '{"root_path":"/media/dracula/hddxfstest/test", "postfix":"1"}' -t '{"root_path":"/media/dracula/hddext4test/test", "postfix":"1"}' -t '{"root_path":"/media/dracula/hddbtrfstest/test", "compress_type":"none"}' -t '{"root_path":"/media/dracula/hddbtrfstest/test", "compress_type":"lzo"}' -t '{"root_path":"/media/dracula/hddbtrfstest/test", "compress_type":"zstd:1"}' -t '{"root_path":"/media/dracula/hddbtrfstest/test", "compress_type":"zstd:3"}'

Here is typical results from 1.4k/2.7GB build folder. btrfs is faster than xfs and ext4 in general. ext4 is the slowest. zstd and lzo has huge advantage than none compressed btrfs. Because here the bottle neck is the disk I/O. Smaller data matter a lot. ZSTD:3 impose some pressure on CPU but does not give much advantage than ZSTD:1.

file system disk usage write time (s) write MiB/s tar to file read MiB/s cp read MiB/s
ext4.1 0.106865 32.4621 82.2824 76.1464 83.3907
xfs.1 0.112776 25.8475 103.339 98.3266 99.6622
ext4.0 0.213728 29.8597 89.4537 72.7791 79.1887
xfs.0 0.217307 25.2035 105.98 97.814 99.3223
btrfs.lzo 0.000147705 15.1674 176.105 193.557 195.507
btrfs.none 0.0505009 22.912 116.579 116.823 118.195
btrfs.zstd-1 0.155232 11.5432 231.397 183.332 251.207
btrfs.zstd-3 0.191588 11.2101 238.273 170.086 297.061

Writing 149k/2GB files to mechanic HDD is extremely slow for xfs and non-compressed btrfs. The speed is as low as 1MB/s for btrfs almost impossible to use. zstd is 100x faster than non-compressed one. ext4 is 5x faster than xfs, 45x faster than btrfs. lzo in this case only have 50% speed of zstd:1. zstd:3 is 10% slower than zstd:1 too.

Reading from it is even slower. btrfs in general is much faser than ext4, the xfs is the slowest only 7MiB/s.

file system disk usage write time (s) write MiB/s tar to file read MiB/s cp read MiB/s
xfs.0 0.112778 171.254 9.51661 8.53608 7.72498
btrfs.zstd-1 0.000147705 13.399 121.632 34.2228 30.4638
ext4.0 0.106865 36.2998 44.8971 7.45604 11.5033
xfs.1 0.192459 189.117 8.6177 8.44728 7.53305
btrfs.none 0.0261714 1394.19 1.16896 12.4196 20.2212
ext4.1 0.186687 69.3157 23.5121 7.20049 11.8862
btrfs.lzo 0.10633 26.8329 60.7373 19.6596 23.9906
btrfs.zstd-3 0.143793 14.9673 108.888 32.3755 22.2033

btrfs is smart enough to skip compression on most uncompressible files

Here is an example, the .git and pngs are ignored in all compression types.

du -h -d1 mongo-cxx-driver
60M     mongo-cxx-driver/.git
96K     mongo-cxx-driver/benchmark
3.9M    mongo-cxx-driver/data
172K    mongo-cxx-driver/etc
4.0M    mongo-cxx-driver/src
60K     mongo-cxx-driver/.evergreen
4.0K    mongo-cxx-driver/build
404K    mongo-cxx-driver/examples
40K     mongo-cxx-driver/cmake
2.0M    mongo-cxx-driver/docs
8.0K    mongo-cxx-driver/generate_uninstall
108K    mongo-cxx-driver/debian
71M     mongo-cxx-driver
du -h -d 1 .
71M     ./mongo-cxx-driver
35M     ./images
541M    ./media
46M     ./pdfs


Processed 1225 files, 1531 regular extents (1531 refs), 657 inline.
Type       Perc     Disk Usage   Uncompressed Referenced  
TOTAL       96%      663M         690M         690M       
none       100%      646M         646M         646M       
lzo         38%       16M          43M          43M          
Processed 1225 files, 1520 regular extents (1520 refs), 669 inline.
Type       Perc     Disk Usage   Uncompressed Referenced  
TOTAL       95%      657M         690M         690M       
none       100%      646M         646M         646M       
zstd        26%       11M          43M          43M       
Processed 1225 files, 1520 regular extents (1520 refs), 669 inline.
Type       Perc     Disk Usage   Uncompressed Referenced  
TOTAL       95%      657M         690M         690M       
none       100%      646M         646M         646M       
zstd        26%       11M          43M          43M
  
in lzo:
for _d in *; do echo $_d; sudo compsize -x $_d; done
images
Processed 95 files, 94 regular extents (94 refs), 1 inline.
Type       Perc     Disk Usage   Uncompressed Referenced  
TOTAL       99%       34M          34M          34M       
none       100%       34M          34M          34M       
lzo         65%      316B         486B         486B       
media
Processed 13 files, 767 regular extents (767 refs), 0 inline.
Type       Perc     Disk Usage   Uncompressed Referenced  
TOTAL       99%      537M         540M         540M       
none       100%      530M         530M         530M       
lzo         70%      7.0M          10M          10M       
mongo-cxx-driver
Processed 1113 files, 467 regular extents (467 refs), 656 inline.
Type       Perc     Disk Usage   Uncompressed Referenced  
TOTAL       92%       63M          69M          69M       
none       100%       59M          59M          59M       
lzo         41%      3.8M         9.2M         9.2M       
pdfs
Processed 4 files, 203 regular extents (203 refs), 0 inline.
Type       Perc     Disk Usage   Uncompressed Referenced  
TOTAL       58%       26M          45M          45M       
none       100%       20M          20M          20M       
lzo         23%      5.7M          24M          24M

in above example, we can see the .git, pngs, pdfs and mp3 are not compressed in all lzo/zstd file systems.

The global compressions have no observable adverse effect on W/R of SATA-SSD.

index file system disk usage write time (s) write MiB/s write rate tar to file read MiB/s tar read rate cp read MiB/s cp read rate
3 xfs.1 0.0325174 2.72032 228.279 0.980201 333.153 0.988391 358.743 1.00002
7 ext4.1 0.0248249 2.70295 229.746 0.9865 336.028 0.996918 356.526 0.993843
2 btrfs.none 0.0236382 2.68499 231.283 0.993099 374.854 1.11211 358.245 0.998634
0 xfs.0 0.00823614 2.63784 235.417 1.01085 336.127 0.997213 363.216 1.01249
6 btrfs.zstd-3 0.0712845 2.63473 235.694 1.01204 382.901 1.13598 371.125 1.03454
5 ext4.0 1.40477e-06 2.63095 236.034 1.0135 338.105 1.00308 360.944 1.00616
1 btrfs.lzo 0.000147705 2.53214 245.244 1.05305 375.027 1.11262 352.94 0.983847
4 btrfs.zstd-1 0.047963 2.51963 246.462 1.05828 372.118 1.10399 354.796 0.989019

The W/R of incompressible files on NVME-SSD are not affected by the global compression.

index file system disk usage write time (s) write MiB/s write rate tar to file read MiB/s tar read rate cp read MiB/s cp read rate
2 ext4.1 1.40477e-06 1.1026 563.207 0.990759 1079.53 0.986251 1244.28 1.00085
3 ext4.0 0.0248249 1.08241 573.713 1.00924 1109.63 1.01375 1242.15 0.999145
0 btrfs.none 0.000147705 0.980839 633.123 1.11375 1142.03 1.04335 1012.4 0.814339
1 xfs.1 0.00760381 0.974274 637.389 1.12126 1225.12 1.11926 1269.08 1.02081
5 btrfs.lzo 0.0244737 0.944111 657.753 1.15708 1112.02 1.01594 1102.44 0.886767
4 xfs.0 0.0197451 0.938866 661.428 1.16354 1207.72 1.10337 1368.09 1.10045
6 btrfs.zstd-3 0.0479703 0.931245 666.84 1.17306 1132.01 1.0342 1101.61 0.886101
7 btrfs.zstd-1 0.0712979 0.910304 682.181 1.20005 1141.54 1.0429 1126.06 0.905768
index file system disk usage write time (s) write MiB/s write rate tar to file read MiB/s tar read rate cp read MiB/s cp read rate
2 ext4.1 1.40477e-06 1.1036 562.696 0.98094 1093.72 0.99911 1210.67 0.983837
3 ext4.0 0.0248249 1.06232 584.563 1.01906 1095.67 1.00089 1250.45 1.01616
0 xfs.0 0.0076032 1.04565 593.88 1.0353 1187.42 1.0847 1333.09 1.08332
7 btrfs.zstd-1 0.0712955 0.971421 639.262 1.11442 1132.66 1.03468 1115.22 0.906273
1 btrfs.zstd-3 0.000147705 0.968957 640.887 1.11725 1051.65 0.960673 1021.85 0.830392
6 btrfs.lzo 0.0478038 0.948783 654.515 1.14101 1148.56 1.0492 1072.48 0.871542
5 btrfs.none 0.0234766 0.940141 660.531 1.15149 1131.13 1.03328 1121.89 0.911694
4 xfs.1 0.0197444 0.92074 674.449 1.17576 1189 1.08615 1338.7 1.08788

Though some times the non-comressed one is a bit faster but it is just random noise.

I only see performance drop caused by compression on HDD.

index file system disk usage write time (s) write MiB/s write rate tar to file read MiB/s tar read rate cp read MiB/s cp read rate
4 ext4.0 0.106865 8.45322 73.4622 0.94311 78.8442 0.988075 84.4509 0.924508
1 btrfs.lzo 0.0234729 7.87219 78.8843 1.01272 114.604 1.43622 97.4419 1.06672
0 btrfs.zstd-3 0.000147705 7.56092 82.1319 1.05441 113.929 1.42776 102.339 1.12033
5 ext4.1 0.131688 7.54319 82.3249 1.05689 80.7473 1.01192 98.2429 1.07549
2 xfs.1 0.112776 7.4823 82.9948 1.06549 91.0122 1.14056 94.2931 1.03225
3 btrfs.zstd-1 0.0469597 7.37862 84.161 1.08046 114.727 1.43775 98.6443 1.07989
6 xfs.0 0.137055 6.92109 89.7247 1.15189 88.1998 1.10532 91.6923 1.00378
7 btrfs.none 0.0702849 5.36696 115.706 1.48544 102.403 1.28331 100.148 1.09635

The non-compressed btrfs stably out-performance the compresed ones by 40% on writing. The reading speed is not affected at all. However, considering the super bad performance of non-compressed btrfs on small files. I still advise to turn on the zstd:1 globally on HDD.

choices

system with powerful CPUs

For normal build dirs:

hardware best 2nd 3rd worst
USB3-SATA-SSD btrfs zstd:3 btrfs zstd:1 btrfs lzo btrfs
NVME-SSD btrfs lzo btrfs zstd:1 btrfs xfs
HDD btrfs zstd:3 btrfs zstd:1 btrfs lzo ext4

For build dirs contain large number of small files:

hardware best 2nd 3rd worst
USB3-SATA-SSD btrfs zstd/lzo btrfs zstd/lzo btrfs zstd/lzo xfs
NVME-SSD btrfs lzo btrfs ext4 btrfs zstd:1
HDD btrfs zstd:1 btrfs zstd:3 btrfs lzo btrfs
  • SATA-SSD: I recommend btrfs zstd:1 or lzo. But any FS seems ok, though XFS is the worst choice.
  • NVME-SSD: we should stick to btrfs lzo. It is way faster than the rest. However, others are still usable.
  • HDD: I recommend btrfs zstd:1. We should definitely avoid btrfs here, because it is almost not usable at all: 1MiB/s in many-small-files case.

system with weaker CPUs

in all tests on internal nvme SSD, the zstd:1 and zstd:3 usually show worst performance: 60-80% of ext4. The lzo in ffmpeg tests show almost identical performance as ext4 but much worse in many-small-files tests.

on slower disk: usbc-5gb-nvme, compressed btrfs are still the fastest ones.

Conclusion

Though the many-small-files case yield very bad performance in general, the SSDs still perform far better than HDD. If I need to work on HDD, I will need to be very careful about where to put small files, definitely not on standard btrfs, though it claims to inline small files^1^. Since the btrfs do not compress uncompressable files, we can turn on the compression globally to avoid the small file issue.

zstd:1 still has huge penalty on I/O speed on modern PC system. The compression is way slower than decompression. The tests on relatively slow SATA SSD show no obvious performance drop because the bottle neck is still disk I/O. On NVME SSD the penalty become far serious. Thus we should only use lzo to reduce the size on NVME-SSD.

In other hand, if the drive is much slower such as HDD, usually only have less than 100MB/s. Then the benefit of smaller data size is far more beneficial than SSDs. Indeed, the degree of benefits completely depends on the information density of the data. If zstd/lzo can yield reasonable compression ratio (<50%) it might worth to turn on compression on slow storage devices.

LZO can give mild compression around 50% for most of my build dir. It requires way less CPU than ZSTD:1. Thus on fast internal SSDs, LZO is preferred.

#!/bin/bash
set -e
set -x
start=1
part_size=5120
parted --script $1 \
mklabel gpt \
mkpart primary ${start}MiB $(($part_size + $start))MiB \
mkpart primary $(($part_size + $start))MiB $(($part_size*2 + $start))MiB \
mkpart primary $(($part_size*2 + $start))MiB $(($part_size*3 + $start))MiB \
mkpart primary $(($part_size*3 + $start))MiB $(($part_size*4 + $start))MiB \
mkpart primary $(($part_size*4 + $start))MiB $(($part_size*5 + $start))MiB \
mkpart primary $(($part_size*5 + $start))MiB $(($part_size*6 + $start))MiB
parted $1 unit MiB print free
fs_types=(
ext4
btrfs
xfs
)
prefix=usb2
keyfile="${prefix}luks.key"
echo -e "test" > "${keyfile}"
for _i in "${!fs_types[@]}"
do
wipefs -a "${1}$(($_i + 1))"
mkfs.${fs_types[$_i]} "${1}$(($_i + 1))" -L "${prefix}${fs_types[$_i]}"
done
for i in {4..6}
do
cryptsetup -v --sector-size=4096 luksFormat "${1}${i}" "${keyfile}"
done
prefix=usb2luks
for i in {4..6}
do
cryptsetup open --type luks "${1}${i}" "${prefix}-$(basename ${1})$i" --key-file="${keyfile}"
done
for _i in "${!fs_types[@]}"
do
mkfs.${fs_types[$_i]} /dev/mapper/"${prefix}-$(basename ${1})$(($_i + 4))" -L "${prefix}${fs_types[$_i]}"
done
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment