Advice on doing several characteristic writes in a row as fast as possible ?

[AKYR]

Hello Nordic team, 

I'm currently working on a project on a nrf52840 DK with the SDK 15.3.0.

As central role, I have an external component that sends big chunks of data to be written in the peripheral device characteristic. 

The chunks size are about 512 bytes and the characteristic to write the data in has a maximum size of 20 bytes. 

To send the data I'm using a function similar to the uart client example :  ble_nus_c_string_send(ble_nus_c_t * p_ble_nus_c, uint8_t * p_string, uint16_t length).

My current idea to implement this is to add a FIFO to buffer the incomming data and do multiple calls to ble_nus_c_string_send to send all the data every time the sd_ble_gattc_write() returns the  BLE_GATTC_EVT_WRITE_CMD_TX_COMPLETE event to empty the FIFO.

Is this the best way to achieve my purpose or do you suggest another way to do this ? I need the data to be sent the fastest it can.

I have seen in the SDK some code about Queued Writes module : https://infocenter.nordicsemi.com/index.jsp?topic=%2Fcom.nordic.infocenter.sdk5.v15.0.0%2Fgroup__nrf__ble__qwr__config.html

Would this be a good way to realize this ? 

Thanks in advance,

Aloïs KYROU.

[Vidar Berg]

Hi Aloïs,

Please have a look at the attached example below. I based it on the ble_app_uart_c example in sdk 15.3.0 and added a ring buffer to queue write commands in the app in order to handle any "no resource" error from the send API. 

ble_app_uart_c_mod.zip

Note: the timer instance used to process the buffer in this example is not really necessary, and the project is configured for 52832 so you should copy the source code to a 52840 project before you try it. 

[AKYR]

Ok I see it,

My implementation is kinda similar but I write in a TX_FIFO instead of inserting a tx_buffer and process to send the maximum amount of data I can from the TX_FIFO when you are processing a buffer. 

I guess both method work :) 

Do you have an argument to prefer one over the other ? 

The only one I can see is in the case of small data inputs you will create a lot of buffer of small size whereas the data will br write in queue in the FIFO (but it is not my case so...). 

Thank you !

[Vidar Berg]

Yes, both approaches should work, although mine is not optimized for RAM usage:) I used a ring buffer because it allowed me to reuse the buffer implementation from the ble_app_ancs_c example.