why fatfs f_write has unstable speed?

[Saghi.1]

Hi

i'm using an own made stm32h750vb board to write on sd memory using fatfs third party lib from cubemx.

the problem i encountered is unstable function of f_write. sometimes it takes more time to write.

to measure the time f_write takes i set a gpio pin before and after it, and monitored ​the pin by oscilloscope.

in my code i generate a timer interrupt to write on certain times on sd memory and compare to gpio pin i set to measure f_write speed.

in picture up signal shows gpio pin set before f_write and reset after f_write, and down signal edges show timer interrupt that f_write starts on this transition.​

as you see ​at first picture(left one) sometimes f_write takes more time. result is for writing a 16k block size dummy data, 400mhz cpu clock, 50mhz sdmmc clock with 4bit interface,and a sdhc microsd 4gig c4 memory.

i tried different data block size(512byte,1Kb,4Kb,8Kb,32Kb) with different swmmc clock speed from 25mhz to 100mhz), but the problem still exists.

i tried to investigate if the problem is with f_wrie function or bsp_sd_driver​. i put the gpio test pin before and after hal_write_sd_blocks function that interfacing with sdmmc in lower level of f_write and it seems to work fine (secondpic) and inconsistency of write to sd is related to something in f_write function!

please help.​

[berendi]

It is caused by the internal controller in the card, doing wear leveling, ECC recovery and whatnot. You can't do anything about it. An SD card is a consumer product not suitable for real time operation.

Delays of 10 to 20 ms are quite common, up to 250 ms have been observed. Get a bigger buffer for your data, write in a separate task, and pray.

[Tesla DeLorean]

FatFs manages file structures and Fat tables, your f_write may cause several read and write operations at the DISKIO layer, instrument that to better understand the interactions. The command/response of the media's controller can be relatively slow at times.​

[Tesla DeLorean]

All the source code is available for inspection​.

Check buffer alignment, and read the cards erase block size info, it is doing a lot of shuffling of block resources​ on the NAND.

FatFs manages the blocking/deblocking or unaligned writes into a file.​

[Saghi.1]

thanks berendy,I admit your answer somehow, but i doubt that.

i tried a different sd memories,write delay varied but still exists. also as you see in picture some delays of writing is not related to hal_sd_write_blocks that write data to sd and seems to be related to f_write function operation!!

[Saghi.1]

so you are approving that problem is related to sd memory operation?

is there any class or type of sd memory that have less delay or inconsistency?​

[TDK]

> is there any class or type of sd memory that have less delay or inconsistency?​

Faster cards will be more responsive, but no cards will be instantaneous for all operations, particular with file system operations which can write to several areas for a single file operation. The delay can also vary depending on factors outside of your control. That's just how SD works. You're not interfacing with the memory directly.

[berendi]

Look for cards with A2 speed rating.

[Tesla DeLorean]

I'm saying a couple of things, a) interactions are more complex than just your requests to f_write(), b) the cards ultimately pace the access speeds and their internal buffering/blocking strategy.

High end brand devices will have better and more consistent performance than "catch-of-the-day" devices

[Saghi.1]

thanks to all.

i got the fastest sd memory supported by stm32h7 series that is a sdhc uhs-i U3 sd memory and it seems the problem solved. this class of sd memory guarantee for at least 30MB/sec speed.​