TL; DR: This article talks about a convenient, fast and reliable way to identify Linux programs that write data to disk, which helps in identifying a large or abnormally frequent load on the disk subsystem, and also allows you to estimate the overhead of the file system. This is especially true for SSDs in PCs, EMMCs and Flash memory in single board computers.
In the course of writing the article, it was found that writing a few kilobytes of data to the BTRFS file system leads to writing 3 megabytes of real data to disk.
Introduction
βOh, nonsense, the memory cells on modern SSDs will fail after decades of normal use, you should not worry about it, much less move swap, virtual machines and the browser profile folder to the HDDβ - a typical answer to the question about the reliability of solid state drives with guaranteed β150 TBW. If you estimate how much typical software can write data, it seems that 10-20 GB per day is already a big figure, let it be a maximum of 40 GB, much more. With such figures, the answer is quite reasonable - it takes 10 years to achieve guaranteed values ββby the number of cell overwrites, with 40 GB of recorded data daily.
However, in 6 years I have already been using the third SSD: the first one had a controller failure, and the second one began to move data between cells several times a day, which resulted in 30-second delays in write maintenance.
After 7 months of using the new SSD, I decided to check the amount of data written, as reported by the drive itself via SMART.
19.7 TB.
In just 7 months, I used 13% of the guaranteed amount of data written, despite the fact that it is configured in accordance with the recommendations for aligning partitions and setting up the FS, I almost never use swap, virtual machine disks are placed on the HDD!
This is an anomalously high figure, at this rate the warranty TBW will be exceeded before reaching the drive's 5-year warranty period. And my computer cannot write 93 gigabytes a day! We need to check how much data is written to disk in 10 minutes ...
Total:
Writes Queued: 24,712, 2,237MiB
Writes Completed: 25,507, 2,237MiB
Write Merges: 58, 5,472KiB
2.2 GiB, wow!
Determining the amount of data written to a disk device
If your device supports SMART (SSD, EMMC, some industrial MicroSD), then the first step is to request data from the drive with programs smartctl, skdump or mmc (from mmc-utils).
Sample output from smartctl program
$ sudo smartctl -a /dev/sdb
smartctl 7.0 2019-03-31 r4903 [x86_64-linux-5.3.11-200.fc30.x86_64] (local build)
Copyright (C) 2002-18, Bruce Allen, Christian Franke, www.smartmontools.org
=== START OF INFORMATION SECTION ===
Model Family: Samsung based SSDs
Device Model: Samsung SSD 860 EVO mSATA 250GB
Serial Number: S41MNC0KA13477K
LU WWN Device Id: 5 002538 e700fa64b
Firmware Version: RVT41B6Q
User Capacity: 250β―059β―350β―016 bytes [250 GB]
Sector Size: 512 bytes logical/physical
Rotation Rate: Solid State Device
Form Factor: mSATA
Device is: In smartctl database [for details use: -P show]
ATA Version is: ACS-4 T13/BSR INCITS 529 revision 5
SATA Version is: SATA 3.1, 6.0 Gb/s (current: 3.0 Gb/s)
Local Time is: Tue Nov 19 01:48:50 2019 MSK
SMART support is: Available - device has SMART capability.
SMART support is: Enabled
=== START OF READ SMART DATA SECTION ===
SMART overall-health self-assessment test result: PASSED
General SMART Values:
Offline data collection status: (0x00) Offline data collection activity
was never started.
Auto Offline Data Collection: Disabled.
Self-test execution status: ( 0) The previous self-test routine completed
without error or no self-test has ever
been run.
Total time to complete Offline
data collection: ( 0) seconds.
Offline data collection
capabilities: (0x53) SMART execute Offline immediate.
Auto Offline data collection on/off support.
Suspend Offline collection upon new
command.
No Offline surface scan supported.
Self-test supported.
No Conveyance Self-test supported.
Selective Self-test supported.
SMART capabilities: (0x0003) Saves SMART data before entering
power-saving mode.
Supports SMART auto save timer.
Error logging capability: (0x01) Error logging supported.
General Purpose Logging supported.
Short self-test routine
recommended polling time: ( 2) minutes.
Extended self-test routine
recommended polling time: ( 85) minutes.
SCT capabilities: (0x003d) SCT Status supported.
SCT Error Recovery Control supported.
SCT Feature Control supported.
SCT Data Table supported.
SMART Attributes Data Structure revision number: 1
Vendor Specific SMART Attributes with Thresholds:
ID# ATTRIBUTE_NAME FLAG VALUE WORST THRESH TYPE UPDATED WHEN_FAILED RAW_VALUE
5 Reallocated_Sector_Ct 0x0033 100 100 010 Pre-fail Always - 0
9 Power_On_Hours 0x0032 098 098 000 Old_age Always - 5171
12 Power_Cycle_Count 0x0032 099 099 000 Old_age Always - 459
177 Wear_Leveling_Count 0x0013 096 096 000 Pre-fail Always - 62
179 Used_Rsvd_Blk_Cnt_Tot 0x0013 100 100 010 Pre-fail Always - 0
181 Program_Fail_Cnt_Total 0x0032 100 100 010 Old_age Always - 0
182 Erase_Fail_Count_Total 0x0032 100 100 010 Old_age Always - 0
183 Runtime_Bad_Block 0x0013 100 100 010 Pre-fail Always - 0
187 Uncorrectable_Error_Cnt 0x0032 100 100 000 Old_age Always - 0
190 Airflow_Temperature_Cel 0x0032 058 039 000 Old_age Always - 42
195 ECC_Error_Rate 0x001a 200 200 000 Old_age Always - 0
199 CRC_Error_Count 0x003e 100 100 000 Old_age Always - 0
235 POR_Recovery_Count 0x0012 099 099 000 Old_age Always - 29
241 Total_LBAs_Written 0x0032 099 099 000 Old_age Always - 38615215765
SMART Error Log Version: 1
No Errors Logged
SMART Self-test log structure revision number 1
No self-tests have been logged. [To run self-tests, use: smartctl -t]
SMART Selective self-test log data structure revision number 1
SPAN MIN_LBA MAX_LBA CURRENT_TEST_STATUS
1 0 0 Not_testing
2 0 0 Not_testing
3 0 0 Not_testing
4 0 0 Not_testing
5 0 0 Not_testing
Selective self-test flags (0x0):
After scanning selected spans, do NOT read-scan remainder of disk.
If Selective self-test is pending on power-up, resume after 0 minute delay.My SSD stores the amount of data written in parameter 241 Total_LBAs_Written, in logical blocks (LBAs), not bytes. The size of the logical block in my case is 512 bytes (you can see it in the output of smartctl, in Sector Size). To get bytes, you need to multiply the parameter value by 512.
38615215765 Γ 512 Γ· 1000 Γ· 1000 Γ· 1000 Γ· 1000 = 19,770 Π’Π
38615215765 Γ 512 Γ· 1024 Γ· 1024 Γ· 1024 Γ· 1024 = 17,981 Π’ΠΈΠProgram skdump on my SSD is trying to interpret the value of Total_LBAs_Written somehow in its own way, because of which it displays 1296217.695 TB, which is obviously incorrect.
To find out the amount of recorded information at the device level, we will use the program btrace from the package blktrace. It shows both general statistics for the entire time the program was running, as well as individual processes and threads (including kernels) that performed the recording.
Run the following command to gather information in 10 minutes, where /dev/sdb is your drive:
# btrace -w 600 -a write /dev/sdbTypical command output
β¦
8,16 0 3253 50.085433192 0 C WS 125424240 + 64 [0]
8,16 0 3254 50.085550024 0 C WS 193577744 + 64 [0]
8,16 0 3255 50.085685165 0 C WS 197246976 + 64 [0]
8,16 0 3256 50.085936852 0 C WS 125736264 + 128 [0]
8,16 0 3257 50.086060780 0 C WS 96261752 + 64 [0]
8,16 0 3258 50.086195031 0 C WS 94948640 + 64 [0]
8,16 0 3259 50.086327355 0 C WS 124656144 + 64 [0]
8,16 0 3260 50.086843733 15368 C WSM 310218496 + 32 [0]
8,16 0 3261 50.086975238 753 A WSM 310218368 + 32 <- (8,20) 291339904
8,16 0 3262 50.086975560 753 Q WSM 310218368 + 32 [dmcrypt_write/2]
8,16 0 3263 50.086977345 753 G WSM 310218368 + 32 [dmcrypt_write/2]
8,16 0 3264 50.086978072 753 I WSM 310218368 + 32 [dmcrypt_write/2]
8,16 0 3265 50.086979159 753 D WSM 310218368 + 32 [dmcrypt_write/2]
8,16 0 3266 50.087055685 0 C WSM 310218368 + 32 [0]
8,16 0 3267 50.087060168 753 A WSM 310218592 + 160 <- (8,20) 291340128
8,16 0 3268 50.087060367 753 Q WSM 310218592 + 160 [dmcrypt_write/2]
8,16 0 3269 50.087061242 753 G WSM 310218592 + 160 [dmcrypt_write/2]
8,16 0 3270 50.087061698 753 I WSM 310218592 + 160 [dmcrypt_write/2]
8,16 0 3271 50.087062361 753 D WSM 310218592 + 160 [dmcrypt_write/2]
8,16 0 3272 50.087386179 0 C WSM 310218592 + 160 [0]
8,16 0 3273 50.087436417 15368 A FWS 0 + 0 <- (253,1) 0
8,16 0 3274 50.087437471 15368 Q FWS [LS Thread]
8,16 0 3275 50.087440862 15368 G FWS [LS Thread]
8,16 0 3276 50.088300047 0 C WS 0 [0]
8,16 0 3277 50.088470917 753 A WFSM 18882688 + 8 <- (8,20) 4224
8,16 0 3278 50.088471091 753 Q WFSM 18882688 + 8 [dmcrypt_write/2]
8,16 0 3279 50.088471688 753 G WFSM 18882688 + 8 [dmcrypt_write/2]
8,16 0 3280 50.088474334 32254 D WSM 18882688 + 8 [kworker/0:2H]
8,16 0 3281 50.088515572 0 C WSM 18882688 + 8 [0]
8,16 0 3282 50.089229069 0 C WSM 18882688 [0]
CPU0 (8,16):
Reads Queued: 0, 0KiB Writes Queued: 345, 25,932KiB
Read Dispatches: 0, 0KiB Write Dispatches: 331, 25,788KiB
Reads Requeued: 0 Writes Requeued: 0
Reads Completed: 0, 0KiB Writes Completed: 1,597, 117,112KiB
Read Merges: 0, 0KiB Write Merges: 1, 16KiB
Read depth: 0 Write depth: 177
IO unplugs: 0 Timer unplugs: 0
CPU1 (8,16):
Reads Queued: 0, 0KiB Writes Queued: 502, 39,948KiB
Read Dispatches: 0, 0KiB Write Dispatches: 495, 40,076KiB
Reads Requeued: 0 Writes Requeued: 0
Reads Completed: 0, 0KiB Writes Completed: 0, 0KiB
Read Merges: 0, 0KiB Write Merges: 0, 0KiB
Read depth: 0 Write depth: 177
IO unplugs: 0 Timer unplugs: 0
CPU2 (8,16):
Reads Queued: 0, 0KiB Writes Queued: 297, 26,800KiB
Read Dispatches: 0, 0KiB Write Dispatches: 287, 26,800KiB
Reads Requeued: 0 Writes Requeued: 0
Reads Completed: 0, 0KiB Writes Completed: 0, 0KiB
Read Merges: 0, 0KiB Write Merges: 0, 0KiB
Read depth: 0 Write depth: 177
IO unplugs: 0 Timer unplugs: 0
CPU3 (8,16):
Reads Queued: 0, 0KiB Writes Queued: 418, 24,432KiB
Read Dispatches: 0, 0KiB Write Dispatches: 408, 24,448KiB
Reads Requeued: 0 Writes Requeued: 0
Reads Completed: 0, 0KiB Writes Completed: 0, 0KiB
Read Merges: 0, 0KiB Write Merges: 2, 272KiB
Read depth: 0 Write depth: 177
IO unplugs: 0 Timer unplugs: 0
Total (8,16):
Reads Queued: 0, 0KiB Writes Queued: 1,562, 117,112KiB
Read Dispatches: 0, 0KiB Write Dispatches: 1,521, 117,112KiB
Reads Requeued: 0 Writes Requeued: 0
Reads Completed: 0, 0KiB Writes Completed: 1,597, 117,112KiB
Read Merges: 0, 0KiB Write Merges: 3, 288KiB
IO unplugs: 0 Timer unplugs: 0
Throughput (R/W): 0KiB/s / 2,338KiB/s
Events (8,16): 9,287 entries
Skips: 0 forward (0 - 0.0%)
btrace allows you to visually see the real amount of recorded data, but it is difficult to understand which programs are recording from its output.
Determining the programs that write to the drive
Program iotop will show the processes writing to the disk and the size of the data written.
The most convenient output is provided by the following parameters:
# iotop -obPatSample program output
02:55:47 Total DISK READ : 0.00 B/s | Total DISK WRITE : 30.65 K/s
02:55:47 Actual DISK READ: 0.00 B/s | Actual DISK WRITE: 0.00 B/s
TIME PID PRIO USER DISK READ DISK WRITE SWAPIN IO COMMAND
b'02:55:47 753 be/4 root 0.00 B 0.00 B 0.00 % 0.04 % [dmcrypt_write/2]'
b'02:55:47 788 be/4 root 72.00 K 18.27 M 0.00 % 0.02 % [btrfs-transacti]'
b'02:55:47 15057 be/4 valdikss 216.00 K 283.05 M 0.00 % 0.01 % firefox'
b'02:55:47 1588 ?dif root 0.00 B 0.00 B 0.00 % 0.00 % Xorg -nolisten tcp -auth /var/run/sddm/{398f030f-9667-4dff-b371-81eaae48dfdf} -background none -noreset -displayfd 18 -seat seat0 vt1'
b'02:55:47 15692 be/4 valdikss 988.00 K 9.41 M 0.00 % 0.00 % python3 /usr/bin/gajim'
b'02:55:47 15730 ?dif valdikss 9.07 M 0.00 B 0.00 % 0.00 % telegram-desktop --'
b'02:55:47 2174 ?dif valdikss 1840.00 K 2.47 M 0.00 % 0.00 % yakuake'
b'02:55:47 19827 be/4 root 16.00 K 896.00 K 0.00 % 0.00 % [kworker/u16:7-events_unbound]'
b'02:55:47 19074 be/4 root 16.00 K 480.00 K 0.00 % 0.00 % [kworker/u16:4-btrfs-endio-write]'
b'02:55:47 19006 be/4 root 16.00 K 1872.00 K 0.00 % 0.00 % [kworker/u16:1-events_unbound]'
b'02:55:47 1429 be/4 root 484.00 K 0.00 B 0.00 % 0.00 % accounts-daemon'
b'02:55:47 15820 be/4 valdikss 312.00 K 0.00 B 0.00 % 0.00 % firefox -contentproc -childID 6 -isForBrowser -prefsLen 7894 -prefMapSize 223880 -parentBuildID 20191022164834 -greomni /usr/lib64/firefox/omni.ja -appomni /usr/lib64/firefox/browser/omni.ja -appdir /usr/lib64/firefox/browser 15057 tab'
b'02:55:47 2125 ?dif valdikss 0.00 B 92.00 K 0.00 % 0.00 % plasmashell'
b'02:55:47 1268 be/3 root 0.00 B 4.00 K 0.00 % 0.00 % auditd'
b'02:55:47 1414 be/4 root 0.00 B 4.00 K 0.00 % 0.00 % sssd_nss --uid 0 --gid 0 --logger=files'
b'02:55:47 15238 be/4 valdikss 0.00 B 4.00 K 0.00 % 0.00 % thunderbird'
b'02:55:47 18605 be/4 root 0.00 B 3.19 M 0.00 % 0.00 % [kworker/u16:0-btrfs-endio-write]'
b'02:55:47 18867 be/4 root 0.00 B 96.00 K 0.00 % 0.00 % [kworker/u16:5-btrfs-endio-meta]'
b'02:55:47 19070 be/4 root 0.00 B 160.00 K 0.00 % 0.00 % [kworker/u16:2-btrfs-freespace-write]'
b'02:55:47 19645 be/4 root 0.00 B 2.17 M 0.00 % 0.00 % [kworker/u16:3-events_unbound]'
b'02:55:47 19982 be/4 root 0.00 B 496.00 K 0.00 % 0.00 % [kworker/u16:6-btrfs-endio-write]'Firefox catches the eye, writing 283 megabytes in a few minutes of running iotop.
Determining which files to write to
Information about the process raping the disk is good, but the paths along which the recording is made is even better.
Let's use the program fatrace, which tracks changes to the file system.
# fatrace -f WSample program output
firefox(15057): CW /home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/moz-extension+++e5c304fb-af40-498a-9ba8-47eb0416e933^userContextId=4294967295/idb/3647222921wleabcEoxlt-eengsairo.sqlite-wal
firefox(15057): CW /home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/moz-extension+++e5c304fb-af40-498a-9ba8-47eb0416e933^userContextId=4294967295/idb/3647222921wleabcEoxlt-eengsairo.sqlite
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/https+++habr.com/ls/data.sqlite-journal
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/https+++habr.com/ls/data.sqlite-journal
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/https+++habr.com/ls/data.sqlite-journal
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/https+++habr.com/ls/data.sqlite-journal
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/https+++habr.com/ls/data.sqlite-journal
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/https+++habr.com/ls/data.sqlite-journal
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/https+++habr.com/ls/data.sqlite-journal
firefox(15057): CW /home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/https+++habr.com/ls/usage-journal
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/https+++habr.com/ls/usage
firefox(15057): CW /home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/https+++habr.com/ls/usage
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/https+++habr.com/ls/data.sqlite-journal
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/https+++habr.com/ls/data.sqlite
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/https+++habr.com/ls/data.sqlite
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/https+++habr.com/ls/data.sqlite
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/webappsstore.sqlite-wal
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/webappsstore.sqlite-wal
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/webappsstore.sqlite-wal
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/webappsstore.sqlite-wal
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/webappsstore.sqlite-wal
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/webappsstore.sqlite-wal
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/webappsstore.sqlite-wal
firefox(15057): CW /home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/https+++habr.com/ls/data.sqlite-journal
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/webappsstore.sqlite
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/webappsstore.sqlite
firefox(15057): W /home/valdikss/.mozilla/firefox/xyf4vqh2.default/webappsstore.sqliteFatrace is unable to show the amount of data written due to the use of fairly simple tracking of the fact that files are accessed through inotify.
The output shows how Habr saves my article in the local storage of the browser while I write it, as well as the Group Speed ββββDial extension, which, as it was discovered using fatrace, reads its data every 30 seconds. It reads, not writes: CW in front of the file says that the file is opened for reading and writing, with the simultaneous creation of the file if it does not exist (openat is called with the O_RDWR|O_CREAT flag), but does not say that any information was actually written to the file.
Just in case, to make sure of this, we will use strace, with a filter on file system calls:
strace -yy -e trace=open,openat,close,write -f -p 15057 2>&1 | grep extensioncommand output
[pid 20352] openat(AT_FDCWD, "/home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/moz-extension+++e5c304fb-af40-498a-9ba8-47eb0416e933^userContextId=4294967295/idb/3647222921wleabcEoxlt-eengsairo.sqlite", O_RDWR|O_CREAT|O_CLOEXEC, 0644) = 153</home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/moz-extension+++e5c304fb-af40-498a-9ba8-47eb0416e933^userContextId=4294967295/idb/3647222921wleabcEoxlt-eengsairo.sqlite>
[pid 20352] read(153</home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/moz-extension+++e5c304fb-af40-498a-9ba8-47eb0416e933^userContextId=4294967295/idb/3647222921wleabcEoxlt-eengsairo.sqlite>, "SQLite format 3 20 22 @ d 23"..., 100) = 100
[pid 20352] read(153</home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/moz-extension+++e5c304fb-af40-498a-9ba8-47eb0416e933^userContextId=4294967295/idb/3647222921wleabcEoxlt-eengsairo.sqlite>, "SQLite format 3 20 22 @ d 23"..., 4096) = 4096
[pid 20352] openat(AT_FDCWD, "/home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/moz-extension+++e5c304fb-af40-498a-9ba8-47eb0416e933^userContextId=4294967295/idb/3647222921wleabcEoxlt-eengsairo.sqlite-wal", O_RDWR|O_CREAT|O_CLOEXEC, 0644) = 166</home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/moz-extension+++e5c304fb-af40-498a-9ba8-47eb0416e933^userContextId=4294967295/idb/3647222921wleabcEoxlt-eengsairo.sqlite-wal>
β¦
[pid 20352] read(54</home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/moz-extension+++e5c304fb-af40-498a-9ba8-47eb0416e933^userContextId=4294967295/idb/3647222921wleabcEoxlt-eengsairo.sqlite>, " r4304364354354364- 4204!4'414" 250 &"..., 4096) = 4096
[pid 20352] read(54</home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/moz-extension+++e5c304fb-af40-498a-9ba8-47eb0416e933^userContextId=4294967295/idb/3647222921wleabcEoxlt-eengsairo.sqlite>, " 136Pt2262504 O24532016:"16.27 r245306>2461t1q370"..., 4096) = 4096
[pid 20352] close(77</home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/moz-extension+++e5c304fb-af40-498a-9ba8-47eb0416e933^userContextId=4294967295/idb/3647222921wleabcEoxlt-eengsairo.sqlite-wal>) = 0
[pid 20352] close(54</home/valdikss/.mozilla/firefox/xyf4vqh2.default/storage/default/moz-extension+++e5c304fb-af40-498a-9ba8-47eb0416e933^userContextId=4294967295/idb/3647222921wleabcEoxlt-eengsairo.sqlite>) = 0
There is no call write(), which indicates that the file is not being written to.
Determining File System Overhead
Big difference in readings iotop ΠΈ btrace prompted the idea to test the file system by manually writing data to a file and monitoring btrace readings.
If you completely exclude writing to disk by booting into systemd emergency mode, and write manually a couple of bytes data to an existing file, btrace to SSD with Btrfs reports a record 3 megabyte real data. A freshly created file system on an 8 GB flash drive writes a minimum of 264 KiB when writing a single byte.
For comparison, writing a couple of bytes to a file on ext4 ends up writing 24 kilobytes of data to disk.
In 2017 Jayashree Mohan, Rohan Kadekodi and Vijay Chidambaram , their results for btrfs and ext4 at 4K writes are consistent with mine.
Conclusion and Conclusion
By the described manipulations it was found out:
- CUPS print daemon writes printer job states frequently to /var/cache/cups every minute. The problem was fixed by cleaning /var/spool/cups (even though there were no print jobs);
- The fact that the database is read every 30 seconds by the Group Speed ββDial extension for Firefox;
- Periodic logging by various performance tracking services on Fedora, resulting in several megabytes of data being written to btrfs: pmcd.service, pmie.service, pmlogger.service;
- Huge amplification when writing a small amount of data when using btrfs.
Conclusion: you should not use btrfs if programs often write a small amount of data (a few kilobytes), otherwise it will turn into megabytes of written data. Especially true for single-board computers with OS on MicroSD.
Source: habr.com