USB speed issue on nRF52840

RE: USB speed issue on nRF52840

Kenneth — Thu, 18 Mar 2021 14:50:19 GMT

This is a rather old case, but can you try to first recreate the speed reported below on Linux or Windows 10? This should isolate if the issue is OS related and/or firmware related.

Kenneth

RE: USB speed issue on nRF52840

Constantin — Thu, 04 Feb 2021 11:22:38 GMT

Furthermore, I've checked the

```static ret_code_t cdc_acm_req_out_data_cb(nrf_drv_usbd_ep_status_t status, void * p_context)```

And there despite the baudrate I've set I am always getting only 9600. If I try to connect with pyserial this callback is running 30-50 times until pyserial gives up and starts the processing. This is why it takes approx. 15 seconds to start the transfer under python.

Again, this was with your standard example.

Any suggestion on that?

Kind regards,

Constantin

RE: USB speed issue on nRF52840

Constantin — Thu, 04 Feb 2021 10:32:30 GMT

Hi,

although it is quiet old thread I would like to jump in.

I am facing a similar problem. I implemented my own application an I am wondering that I cannot get more data through the CDC than 85 kB /s. Therefore, I tried the suggested example and I am getting even less.
My test equipment was:

PCA10056 -> connected to macOSX running CoolEdit Pro / SDK 17.0.0

Terminal was connected with 3 MBaud. Data was captured to Textfile. There was no Data-loss at all.
But still the Terminal only says

Approx. 4.2 MByte per minute => 70 kB / s.

RE: USB speed issue on nRF52840

ovrebekk — Thu, 24 Jan 2019 07:51:36 GMT

Hi Asli

My pleasure, I hope you get the performance you need for your project 🙂

Best regards
Torbjørn

RE: USB speed issue on nRF52840

ASLI — Wed, 23 Jan 2019 14:23:07 GMT

Hi Torbjorn,

Yes you are right. CDC ACM device data rate is definitely higher with Linux too. I measured

~570KBytes/sec for few different data lengths.

Thank you for all the help.

Best Regards,

Asli

RE: USB speed issue on nRF52840

ovrebekk — Wed, 23 Jan 2019 09:51:04 GMT

Hi Asli

Unfortunately neither me or the USB developer have any good theories than this, other than the fact that Windows PC's come in a lot of different flavors with different combinations of hardware, device drivers and configurations.

Also it is clear that the CDC ACM stack in Linux is more efficient in terms of ensuring high data rate.

Best regards
Torbjørn

RE: USB speed issue on nRF52840

ASLI — Tue, 22 Jan 2019 19:58:10 GMT

Hi Torbjorn,

I repeated the CDC_ACM tests on another Windows 10 PC and the speed results are very similar to yours or maybe better for small number of frames.

Not sure what is causing this speed difference on different Windows 10 PCs. Any suggestions on this?

Thanks and Best Regards,

Asli

RE: USB speed issue on nRF52840

ovrebekk — Tue, 22 Jan 2019 16:11:42 GMT

Hi Asli

I still get considerably higher data rates. Testing your code with the frame_counter limit set to 100 I got a total time of less than 22.5ms consistently, indicating a transfer speed of 280KBytes/sec or more.

I also tried 400 and 4096, and got similar speeds (less than a second total for 4096 packets).

I did notice that the byte counter in TeraTerm didn't seem to catch all the bytes unless I changed the buffer to contain all standard characters, but on second thought that is probably just because I forgot to enable the binary option.

Best regards
Torbjørn

RE: USB speed issue on nRF52840

ASLI — Mon, 21 Jan 2019 14:16:43 GMT

Hi,

Correct. When LED_CDC_ACM_TX goes high to low test should be done. And yes, you need to reset the board if you want another test .

Best Regards,

Asli

RE: USB speed issue on nRF52840

ovrebekk — Mon, 21 Jan 2019 14:05:29 GMT

So if I understand your test correctly you just keep button 1 pressed until the limit is reached, and then you have to reset the board if you want to do another test?

Best regards
Torbjørn

RE: USB speed issue on nRF52840

ASLI — Fri, 18 Jan 2019 17:17:06 GMT

Hi Torbjorn,

I repeated your test, sent data to host for about 30 seconds couple of times and I measure ~140 - 160 KBytes/sec not as high as your test result.

I also did some measurements checking the number of frames on Tera Term similar to yours and measured the toggling LED_CDC_ACM_TX time but to set number of frames to 100, 500 or another number I did the below.

I know it's not very accurate but this way I can test the small and exact number of frames.

1. I commented out the LED toggling in main.c line#163

case APP_USBD_CDC_ACM_USER_EVT_TX_DONE:
//bsp_board_led_invert(LED_CDC_ACM_TX);
break;

2. Modified the while loop in main function code as below:

......................................................................................................................

for (int i = 0; i< 64; i++)
{
m_tx_buffer[i] = i;
}
static int frame_counter;

// clear the led (it's opposite because of the LEDS_ACTIVE_STATE 0 in pca10056.h)
bsp_board_led_on(LED_CDC_ACM_TX);

while (true)
{
while (app_usbd_event_queue_process())
{
/* Nothing to do */
}

if(m_send_flag)
{
bsp_board_led_off(LED_CDC_ACM_TX); //set the led
ret = app_usbd_cdc_acm_write(&m_app_cdc_acm, m_tx_buffer, 64);
if (ret == NRF_SUCCESS)
{
++frame_counter;
}
}

if (frame_counter == 4096)
{
bsp_board_led_on(LED_CDC_ACM_TX);//clear the led
break;
}

nrf_cli_process(&m_cli_uart);

............................................................................................................

3. On Tera Term I save a log (only binary option is checked)

4. Change the frame number for different data lengths

I press the button and check the scope to see LED_CDC_ACM_TX go high and low during the transfer of total frame numbers and measure the 'high' time. I also check the Tera Term window to check the total amount of bytes as in the below image and use it for the data rate calculation.

Please let me know if you can reproduce similar results.

Best Regards,

Asli

RE: USB speed issue on nRF52840

ovrebekk — Fri, 18 Jan 2019 09:01:45 GMT

Hi Asli

The toggling of the LED will indicate a packet sent, that is correct, but measuring a single interval is not really enough as there is quite a bit of jitter from packet to packet. For this to work properly you would have to measure the total time taken over a larger number of transfers, and calculate an average. Maybe that is what you did to get the attached numbers?

I tried to test the speed from device to host in a slightly different way. I modified the send function of the example to send 64 bytes on each transfer, but still include a counter in the string:

if(m_send_flag)
{
static int frame_counter;
for(int i = 0; i < 64; i++) m_tx_buffer[i] = '-';

size_t size = sprintf(m_tx_buffer, "Hello USB CDC FA demo: %u\r\n", frame_counter);

ret = app_usbd_cdc_acm_write(&m_app_cdc_acm, m_tx_buffer, 64);
if (ret == NRF_SUCCESS)
{
++frame_counter;
}
}

Then I pressed button 1 on the kit for 30 seconds, and noted in the terminal how many transactions came through.

It ended up at 155909 packets of 64 bytes each, averaging out to about 330KB/sec.

Running the test for a long time doesn't appear to slow down the transfer.

Maybe you can share the changes you did to send 10/100/500/4096 frames and I can try it out?

Alternatively, copy my changes above and try to repeat my test scenario.

Best regards
Torbjørn

RE: USB speed issue on nRF52840

ASLI — Thu, 17 Jan 2019 16:13:42 GMT

Hi Torbjorn,

I've just tested sending a binary file from the host to device using Tera Term 'Send File' with binary option, and I see 290-300 KBytes/sec similar to your test results on TeraTerm file transfer window when I send a large file. I also measured the LED_CDC_ACM_RX toggling time and the data rate was within the range ~290 KBytes/sec.

My test results above in the previous reply are for sending data from device to the host which our project will do data transfer mainly on this direction. As I mentioned before I just could measure the LED_CDC_ACM_TX toggling time as a measure of transfer time for the known binary data with various frame numbers.

Can you please confirm using LED_CDC_ACM_TX (or RX) toggling duration is roughly good enough to measure the data rate and confirm my device -> host data transfer rate results are as expected with developers?

Thank you and Best Regards,

Asli

RE: USB speed issue on nRF52840

ovrebekk — Thu, 17 Jan 2019 15:41:48 GMT

Hi Asli

I was able to repeat the test myself. I simply took the CDC_ACM example from SKD v15.2.0 and disabled logging.

I am getting a quite consistent data rate of 290-300 KBytes/sec, and if I send a large file the transfer rate seems very consistent throughout the file.

I tested it both on my new desktop computer and an older laptop with similar results, running Windows 10 in both cases.

Which version of Windows are you using, and are you able to test on another Windows computer to see if there are any differences?

The developer was unavailable today unfortunately, so I didn't have a chance to discuss your results with him. I will try to get a hold of him tomorrow so I can get his feedback.

Best regards
Torbjørn

RE: USB speed issue on nRF52840

ASLI — Wed, 16 Jan 2019 23:18:06 GMT

Hi Torbjorn,

I tried CDC_ACM device data speed test and I had the below results. I disabled logging and used the example modifying the tx data and it's size, sent known binary data and 64 bytes each time. I also increased the COM port baud rate to highest. I measured the time on LED_CDC_ACM_TX using a scope)

10 frames X 64 bytes => ~242 KBytes/sec

100 frames X 64 bytes => ~220 KBytes/sec

500 frames X 64 bytes => ~153 KBytes/sec

4096 frames X 64 bytes => ~133 KBytes/sec (also measured ~120 KBytes or ~150 KBytes/sec)

It seems like data rate is slowing down as the consecutive number of frames are higher. Also the data rate for the same number of frames might vary a bit.

Please let me know your test results and if you have further description about testing if you think above results are unexpected.

Thanks and Best Regards,

Asli

RE: USB speed issue on nRF52840

ovrebekk — Wed, 16 Jan 2019 13:42:06 GMT

Hi Asli

Reviewing the test results again I see that the only test results from Windows was using the CDC class, and the end result was 215 kB/s.

This was using the method desribed below, using Tera Term to send a binary file over the virtual comport.

Have you tried this test?

I will try to repeat it myself to see if I can reproduce the issue of variability that you discovered. The original testing done by the USB team doesn't include any note of this.

Edit: Another important note is that logging must be disabled and the event queue enabled for optimal performance when running the CDC example. If you can't figure out how just let me know, and I will try to provide more detailed instructions.

Best regards
Torbjørn

RE: USB speed issue on nRF52840

ASLI — Tue, 15 Jan 2019 21:41:10 GMT

Hi again,

I tested the MSC device speed as instructed. Attached are the images for Write and Read speed.

I see a high change (drop) in speed as the block and total data sizes are smaller.

- Is this speed drop expected and why is it happening?

Write Test:

Read Test:

I'm working on CDC_ACM speed test in order to evaluate it . I'll update once I have the results but please update with Windows example/instructions with >200 KBytes/sec data rate if available.

Thank you and Best Regards,

Asli

RE: USB speed issue on nRF52840

ASLI — Tue, 15 Jan 2019 15:28:47 GMT

Hi Torbjorn,

Thank you for the update and suggested speed testing.

I'll repeat the tests and update the results when I'm done but I'm specifically looking for speed test with Windows, where the data rate was measured >200 KBytes/sec with Windows as you wrote about earlier in the thread (please check your replies above)

CDC rate is not high enough for our requirement and the issue I'm observing is the low and varying data rate with Windows as I explained above in the ticket.

As an additional update, when I run my code with libusb on linux I see ~480 KBytes/sec and the rate is not varying much as in Windows (with Windows it's varying a lot and the highest rate I get is ~100 KBytes/sec and mostly ~30 - 70 KBytes/sec). So now I'm wondering whether there is a limitation with Windows no matter I'm using libusb and its drivers?

If you can provide the instructions for the speed test with Windows where the rate is > 200 KBytes/sec that would be very helpful.

Thank you and Best Regards,

Asli

RE: USB speed issue on nRF52840

ovrebekk — Tue, 15 Jan 2019 12:04:06 GMT

Hi Asli

I got the following descriptions for CDC and MSC testing respectively.

CDC:

Install Tera Term
Flash CDC ACM example (logging and lack of event queue decreases performance), connect to COM
Change speed to highest available (921600)
Send a big file using binary option

MSC:

Quick notes how to test MSC transfers

Connect MSC demo to the linux machine or linux virtual machine.
Detect the device using dmesg command

dmesg
Search for our devices, they would look like this:

[ 5089.681017] sd 6:0:0:0: [sdb] 64 512-byte logical blocks: (32.8 kB/32.0 KiB)

[ 5089.689992] sd 6:0:0:1: [sdc] 1048576 512-byte logical blocks: (537 MB/512 MiB)

Test read

sdc is a detected device name:

sudo dd if=/dev/sdc of=/dev/null bs=1M count=32 status=progress

Test write

sudo dd if=/dev/zero of=/dev/sdc bs=1M count=32 status=progress

Real time statistic

The status=progress seems not to work properly with /dev/zero as a source. How to view statistic anyway:

http://askubuntu.com/questions/215505/how-do-you-monitor-the-progress-of-dd

Best regards
Torbjørn

RE: USB speed issue on nRF52840

ovrebekk — Thu, 10 Jan 2019 08:12:01 GMT

Hi Asli

The device side code is just the standard SDK examples.

Currently the USB developers are unavailable, but once they are back I will ask if they can share some details on the host configuration when these tests were performed.

Best regards
Torbjørn

RE: USB speed issue on nRF52840

ASLI — Tue, 08 Jan 2019 15:10:08 GMT

Hi Torbjorn,

Actually I'm defining a MSC device with bulk endpoints, modifying the usbd example which should provide the highest usb speed according to my understanding. My goal is evaluate the maximum usb rate available and show that it can reach ~600-700 kB/sec which is needed for our application.

I went through the SDK examples but not sure how MSC and CDC examples will help me to reach faster speeds.

Is it possible to provide the USB speed measurement example (with both PC and device side code) where you can measure >200KBytes/sec? This would be a great help.

Thanks,

Asli

RE: USB speed issue on nRF52840

ovrebekk — Tue, 08 Jan 2019 14:39:40 GMT

Hi Asli

Would it be an option to use the CDC or MSC USB classes instead?

We have full examples for this in the SDK, and have tested them to work at speeds of > 200KBytes/sec.

Best regards
Torbjørn

RE: USB speed issue on nRF52840

ASLI — Fri, 04 Jan 2019 22:14:15 GMT

Unfortunately I don't see any improvement on speed with sending a packet size < EP1_MAXPACKETSIZE or zero byte packet. I don't have an issue sending and receiving data. It's the speed of the USB transfer varying a lot and low.

Thanks,

RE: USB speed issue on nRF52840

Turbo J — Fri, 04 Jan 2019 14:13:03 GMT

Same issue as in this CDC problem: You must send a packet < EP1_MAXPACKETSIZE in order to properly finish a transaction. You can send zero bytes in a packet.