USB Mass Storage file transfert extremely slow

RE: USB Mass Storage file transfert extremely slow

Håkon Alseth — Wed, 22 Oct 2025 15:43:25 GMT

Hi Benoît,

Glad to hear that you see an improvement as well.

[quote user="Ben_Squar3d"]

Do you have improved speed when setting CONFIG_MASS_STORAGE_STACK_SIZE to at least 5000?

[/quote]

I did try to adjust a bit here and there, and no option seemed to give a higher throughput, as the bottle neck does seem to be the external flash.

I tested on ubuntu 24.04.

If you are testing on Windows, there will be a bit of a USB CDC_ACM throughput difference as compared to linux. See here for similar test vectors:

USBD with CDC ACM class Windows 10 maximum data rate

Slightly newer measurement here:

RE: [nRF52840] USB CDC baud rate

Kind regards,

Håkon

RE: USB Mass Storage file transfert extremely slow

Ben_Squar3d — Tue, 21 Oct 2025 08:12:20 GMT

Hi,

Thank you very much for your quick and accurate answer! Indeed I see an improvement of the performance, but not as much as you.

For the 1mb_random file : ~20s

For the all zero file : ~30s

It is a real improvement (regarding the 49s at the beginning), but I need to estimate the additionnal current consumption to know if this tradeoff is acceptable.

Do you have improved speed when setting CONFIG_MASS_STORAGE_STACK_SIZE to at least 5000?

Thank you very much for your help

Best regards

RE: USB Mass Storage file transfert extremely slow

Håkon Alseth — Mon, 20 Oct 2025 12:11:50 GMT

Hi Benoît,

Transferring 1 MB over USB should ideally be done in approx. ~5 seconds over the USB interface using USBD peripheral (200-300 kB/s max throughput), so I suspect that the delay is in the external flash.

I have two test files, created with dd:

dd if=/dev/urandom of=1mb_random_file.txt bs=1000K count=1
dd if=/dev/zero of=1mb_file.txt bs=1000K count=1

I consistently see a difference between the timing on these devices where all zero's will use approx. 45-46 seconds:

pv 1mb_file.txt | dd of=/media/hkn/5961-0800/1mb_file.txt
1000KiB 0:00:00 [95,2MiB/s] [===========================================================>] 100%            
2000+0 records in
2000+0 records out
1024000 bytes (1,0 MB, 1000 KiB) copied, 45,4911 s, 22,5 kB/s

And a random pattern will use approx. 34 seconds:

pv 1mb_random_file.txt | dd of=/media/hkn/5961-0800/1mb_file.txt
1000KiB 0:00:05 [ 167KiB/s] [===========================================================>] 100%            
2000+0 records in
2000+0 records out
1024000 bytes (1,0 MB, 1000 KiB) copied, 27,5651 s, 37,1 kB/s

So there is a dependency on writing bitwise '1' vs writing '0'.

What you could try is to set it up in a "high performance mode":

&mx25r64 {
	mxicy,mx25r-power-mode = "high-performance";
	partitions {
		compatible = "fixed-partitions";
		#address-cells = <1>;
		#size-cells = <1>;

		storage_partition: partition@0 {
			label = "storage";
			reg = <0x00000000 0x00100000>;
		};
	};
};

I believe the default boot up mode is a ultra-low power mode.

Using this high performance mode of the mx25r64, I see these numbers:

pv 1mb_ones_file.txt | dd of=/media/hkn/5961-0800/1mb_file.txt
1000KiB 0:00:00 [84,5MiB/s] [===========================================================>] 100%            
2000+0 records in
2000+0 records out
1024000 bytes (1,0 MB, 1000 KiB) copied, 15,6999 s, 65,2 kB/s
pv 1mb_file.txt | dd of=/media/hkn/5961-0800/1mb_file.txt
1000KiB 0:00:00 [98,6MiB/s] [===========================================================>] 100%            
2000+0 records in
2000+0 records out
1024000 bytes (1,0 MB, 1000 KiB) copied, 20,7949 s, 49,2 kB/s

Could you check if adding this mode also increases the performance to approx. 15-20 seconds at your end as well?

Here's my test files:

devzone.nordicsemi.com/.../1mb_5F00_random_5F00_file.txt

devzone.nordicsemi.com/.../1mb_5F00_ones_5F00_file.txt

devzone.nordicsemi.com/.../1mb_5F00_file.txt

Kind regards,

Håkon