Why are notifications slower than indications with SoftDevice s130 on nRF51822?

RE: Why are notifications slower than indications with SoftDevice s130 on nRF51822?

run_ar — Fri, 05 Feb 2021 08:00:09 GMT

I understand. Let us know if you have any further questions.

RE: Why are notifications slower than indications with SoftDevice s130 on nRF51822?

brianreinhold — Thu, 04 Feb 2021 09:26:51 GMT

At the moment I am stuck on the nRF51822 as that is what is being used on the production device and s130 is the best we got (they used to use s110 but I was able to upgrade that). s110 had issues with the flash writes (it didn't write) that simply moving to s130 solved without code changes.

RE: Why are notifications slower than indications with SoftDevice s130 on nRF51822?

run_ar — Thu, 04 Feb 2021 08:44:40 GMT

Yes, it's always hard to document these things in a good way. I usually look at the message sequence charts for the SD to see what is going on. For the S130 this is a bit limited: GATTS Handle Value Notification. But we have more details for newer softdevices: GATTS Handle Value Notification (Newer message sequence charts are not necessarily valid for older SD versions)

RE: Why are notifications slower than indications with SoftDevice s130 on nRF51822?

brianreinhold — Tue, 02 Feb 2021 11:28:11 GMT

Thanks. I am well aware of the documentation. But it is not always clear on these critical points. My understanding is now that the NRF_ERROR_BUSY' is ONLY for indications. What was not clear when the BLE_ERROR_NO_TX_PACKETS error came was if the packet was internally queued but before you send another you have to wait for the event. I have noted that the 'count' returned with the event is most frequently one. So I guess there is an attempt to make a one-to-one correspondence between the sd_ble_gatts_hvx call and the event but that is not always possible.

It would help if there was more in the documentation about how indications and notifications work in SoftDevice as that does influence code design. Its also pretty central in data transfer. I understand its impossible to clarify everything, but in my case I came away with more questions than answers after reading the documentation. I can also understand that sd_* documentation has lower priority as most are probably using the SDK and they are hidden from this stuff.

But by far and away, the best thing about Nordic is the responsiveness they have on this forum!

RE: Why are notifications slower than indications with SoftDevice s130 on nRF51822?

run_ar — Tue, 02 Feb 2021 09:27:42 GMT

For reference: sd_ble_gatts_hvx

1. Most likely you would only see this for indications.

2. Yes, if you get this error it means there was no available buffers, so you need to resend that particular packet.

RE: Why are notifications slower than indications with SoftDevice s130 on nRF51822?

brianreinhold — Sun, 31 Jan 2021 15:20:07 GMT

Okay, didn't know that was how it worked. I assumed that was just a queue and under the hood it would send and generate a response event for each packet in the queue. But if one can stuff 6 packets in a connection interval and that generates only one event, then its an entirely different story! I guess I'll find out! Thanks for the suggestion.

I did that and it sure made a difference! Thanks for pointing that out for me. Had no idea it worked like that!

There is NOT a one to one correspondence between packets and events as I assumed. There are far fewer events than packets! However, I did note that the event has a parameter indicating how many packets were sent so you still know if everything you queued at least went out the door. This statement in the documentation led me astray

It is important to note that a notification will consume an application buffer, and will therefore generate a BLE_EVT_TX_COMPLETE event when the packet has been transmitted.

I still have some questions:

1. Can I get an NRF_ERROR_BUSY when doing notifications?

2. If I get a BLE_ERROR_NO_TX_PACKETS response do I have to resend the packet after I wait for the event?

RE: Why are notifications slower than indications with SoftDevice s130 on nRF51822?

Dmitry — Sun, 31 Jan 2021 12:52:03 GMT

Hi,

what I can see from debug log - the time is exactly the same for notifications and indications (97-98 ms), except for the first packet that's not in sync with connection timeline.

[quote userid="83646" url="~/f/nordic-q-a/71103/why-are-notifications-slower-than-indications-with-softdevice-s130-on-nrf51822"]All events handled app checks send data flag. It is set, app calls sd_ble_gatts_hvx() to send first chunk and clears flag[/quote]

Then you have exactly one packet per connection interval, no matter whether it's a notification or indication. Try to change your algorithm to call sd_ble_gatts_hvx() until it returns BLE_ERROR_NO_TX_PACKETS, to fill notification queue and allow SoftDevice to send as many packets in a connection event as it can.