Writing to internal flash with LittleFS (v2) is extremely slow. #17631

saraverbeecke · 2025-07-07T17:08:06Z

saraverbeecke
Jul 7, 2025

When using Micropython 1.25 on a stm32h743 nucleo board with LittleFS (v2). Writing a file to the internal flash seems to be very slow... it takes minimum around 2.5s to write a few hundred bytes.
Any ideas to speed up the writing process?

Thanks!

import time
start = time.ticks_ms()
with open("/flash/test.bin", "wb") as f:
    f.write(b"\x00" * 256)
print("raw write: {:} ms".format(time.ticks_diff(time.ticks_ms(), start)))

256
raw write: 2704 ms

import time
start = time.ticks_ms()
with open("/flash/test.bin", "wb") as f:
    f.write(b"\x00" * 512)
print("raw write: {:} ms".format(time.ticks_diff(time.ticks_ms(), start)))

512
raw write: 2707 ms

import time
start = time.ticks_ms()
with open("/flash/test.bin", "wb") as f:
    f.write(b"\x00" * 1024)
print("raw write: {:} ms".format(time.ticks_diff(time.ticks_ms(), start)))

1024
raw write: 2727 ms

import time
start = time.ticks_ms()
with open("/flash/test.bin", "wb") as f:
    f.write(b"\x00" * 4096)
print("raw write: {:} ms".format(time.ticks_diff(time.ticks_ms(), start)))

4096
raw write: 2710 ms

import time
start = time.ticks_ms()
with open("/flash/test.bin", "wb") as f:
    f.write(b"\x00" * 16384)
print("raw write: {:} ms".format(time.ticks_diff(time.ticks_ms(), start)))

16384
raw write: 2678 ms

import time
start = time.ticks_ms()
with open("/flash/test.bin", "wb") as f:
    f.write(b"\x00" * 65536)
print("raw write: {:} ms".format(time.ticks_diff(time.ticks_ms(), start)))

65536
raw write: 2534 ms

peterhinch · 2025-07-07T17:27:27Z

peterhinch
Jul 7, 2025
Collaborator

On a Pyboard 1.1 the first test runs in 193ms, the last 462ms. I tested using mpremote. Are you using an IDE?
[EDIT]
For the record. When I performed this test the flash was formatted FAT. I repeated the tests with littlefs V2: times were 191ms and 386ms.

9 replies

saraverbeecke Jul 8, 2025
Author

Thanks for your reply. Running your script like you do with rshell gives me the following output. Anything weird here?

Littlefs v2 filesystem
0008  magic                            littlefs
0014  version                          0x20001
0018  block_size                       512
001c  block_count                      1024
0020  name_max                         255
0024  file_max                         2147483647
0028  attr_max                         1022

davefes Jul 8, 2025

After I changed to LFS2 the BlackPill would not load big files, so my comment could be invalid. This board could be damaged,. I have already had to remove the 8MB flash chip.

I get:

Littlefs v2 filesystem
0008  magic                            littlefs
0014  version                          0x20001
0018  block_size                       512
001c  block_count                      128
0020  name_max                         255
0024  file_max                         2147483647
0028  attr_max                         1022

It looks like you have more blocks, the STM32F411 is a relatively small device.

davefes Jul 8, 2025

Just tried again and now the WeACT BlackPill will load big files. Ran your script again and got 182ms.

Totally unrelated but I have found these boards are "quite sensitive" in getting big files loaded consistently, evidently due to the frequency tolerance of the 25MHz system crystal.

saraverbeecke Jul 8, 2025
Author

I am using the stm32h743. And reading files is fast. The writing speed is really weird. Small files (128 bytes) take up 2 seconds.

davefes Jul 8, 2025

I also run Micropython on a STM32H723, which is similar to your board. I haven't got access to it for a week or so. If I had it I would try running your script on it.

You need an expert!

andrewleech · 2025-07-11T22:11:11Z

andrewleech
Jul 11, 2025
Collaborator Sponsor

@saraverbeecke is this test running on a freshly formatted filesystem, or have you been using the board for a while already?

LFS2 can get fragmented which can slow it down significantly.

Also if the filesystem starts to fill up lfs gets really slow, I generally try to ensure my lfs filesystems have more than 20% free otherwise they get incredibly slow.

Oh I see your other comment now saying you did just format it.

I'm not sure what the flash page size is in that micro but it might be quite large, meaning lots needs to be cached into ram to be able to make small changes.

2 replies

saraverbeecke Jul 15, 2025
Author

I ran it indeed on a freshly formatted littlefs2 system. Reading is super fast, writing super slow...
Interesting to read about the caching, can I specify something in mpconfigboard.h? Is the following it?
#define MICROPY_HW_INTERNAL_FLASH_STORAGE_RAM_CACHE (1)

saraverbeecke Jul 15, 2025
Author

ow and btw you mentioned: "LFS2 can get fragmented", is there a way to defragment it somehow without losing data?

MicroPython

Writing to internal flash with LittleFS (v2) is extremely slow. #17631

Uh oh!

saraverbeecke Jul 7, 2025

Replies: 2 comments · 11 replies

Uh oh!

Uh oh!

peterhinch Jul 7, 2025 Collaborator

Uh oh!

saraverbeecke Jul 8, 2025 Author

Uh oh!

davefes Jul 8, 2025

Uh oh!

davefes Jul 8, 2025

Uh oh!

saraverbeecke Jul 8, 2025 Author

Uh oh!

davefes Jul 8, 2025

Uh oh!

Uh oh!

andrewleech Jul 11, 2025 Collaborator Sponsor

Uh oh!

saraverbeecke Jul 15, 2025 Author

Uh oh!

saraverbeecke Jul 15, 2025 Author

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

saraverbeecke
Jul 7, 2025

Replies: 2 comments 11 replies

peterhinch
Jul 7, 2025
Collaborator

saraverbeecke Jul 8, 2025
Author

saraverbeecke Jul 8, 2025
Author

andrewleech
Jul 11, 2025
Collaborator Sponsor

saraverbeecke Jul 15, 2025
Author

saraverbeecke Jul 15, 2025
Author