What is the maximum data transfer speed between DKs?

[Xhimself805]

Hello! I am building a multiple CLI Thread Network. Each CLI Node reads IMU data and send UDP packets out. Each of them keeps doing this. And the nRF5340 DK CLIs connect to the same OTBR (RPi4B + nRF52840DK). But the data transfer meets an bottleneck: when more than two CLIs are connected. There would be a 5000ms latency which is so high. I assume that the data goes in this way: CLI -> RCP -> RPi -> PC. The RPi and PC are so powerful to handle thousands of messages per second. So I suppose the bottleneck is the communication between CLI to RCP (IEEE 802.15.4 )or RCP to RPi (Spinel Protocol, USB CDC?). I use Wireshark on my PC to capture the packets. It seems that when only one CLI is connected, the transfer rate is ~34 kbps. However, as I scale up the number of devices connected to the same UDP port from one to four, the total data rate seemed to saturate at ~68 kbps. This means that once more than two devices are connected, the devices transmit at significantly lower rates than the ~34 kbps seen with just one device to seemingly accommodate for the cap at ~68 kbps. I know the IEEE 802.15.4 has a rate of 250 kbps and the Spinel protocol is 460800 baudrate by default. Where is the bottleneck and how could I solve it? Thank you in advance for your assisstance!

[zhanglongxia]

@Xhimself-Innerfree You can refer to the Thread Quality Dashboard to get the general Thread performance.

I think there are two ways to increase the throughput:
<1> Increase the baud rate of the UART interface to a higher value.
<2> Check if there are other 2.4GHz radios (such as WiFi and BT) nearby. If yes, turn off these devices and try again.

[Xhimself805]

Thank you for your prompt reply @zhanglongxia! I have a quick overview of it, but it seems that many link it is redirecting to has changed. such as the "The summary of throughput result is found on the main Performance Dashboard page." on Page 13. In the link, it doesn't have the summary of throughput result but a testing guide. I will test the throughput of my local setup.
<1>the transfer rate of RCP and CLI are based on IEEE 802.15.4 (250kbps). Can it be calculated like 250000 < 460800? so the bottle neck is between RCP and CLI rather than the Spinel USB CDC?
<2> I would like to turn off those devices and try it again.
Thank you for your assistance, Have a pleasant day!

[jwhui]

5 seconds of latency is too high - there is probably something wrong with your setup.

Espressif recently published a test report, which can be a useful reference for what is achievable.

Some other notes on the limitations of mesh networking (not specific to Thread):

Multihop mesh networking with a half-duplex radio typically observes much less throughput that the base PHY data rate.

Consider the 2-hop scenario where nodes A -> B -> C.

1. A transmits a packet destined for C to B.
2. B forwards the packet to C. During this, B is actively transmitting and cannot actively receive anything from A.

In this case, the theoretical throughput is one-half.

Now consider the 3-hop scenarios where A -> B -> C -> D.

1. A transmits a packet destined for D to B.
2. B forwards the packet to C. During this, B is actively transmitting and cannot actively receive anything from A.
3. C forwards the packet to D. During this, C is actively transmitting and A cannot reliably transmit to B because there would be interference from C.

In this case, the theoretical throughput is one-third.

The above just considers the raw PHY rate. Now add in the overhead for headers, media access, etc and the achievable data rate is even less.

[aghuang1]

Hi @jwhui, I had a follow-up question regarding the test report you shared -- do you happen to know if the reported UDP throughput of 103 kbps for 1-hop communication accounts for the overhead or just the payload throughput?

[jwhui]

The white paper notes that iPerf was used to measure TCP and UDP throughput. iPerf measures and reports the payload data only. It does not include the overhead of network headers such as: MAC header, IP header, and UDP header.

[aghuang1]

Thank you for the reply! I have one quick follow up -- does that mean that if we factor in overhead of network headers, would the total data transfer we should see be closer to 250 kbps (i.e., greater than the 103 kbps for 1-hop communication reported from iPerf)?

[jwhui]

Yes.