Low TCP/UDP throughput on the ESP32 ESP-Hosted-MCU slave (EHM-115)

[nkp123]

Checklist

• Checked the issue tracker for similar issues to ensure this is not a duplicate.
• Provided a clear description of your suggestion.
• Included any relevant context or examples.

Issue or Suggestion Description

Hi,
I am working on connecting ESP32-P4 SoC (ESP32P4NRW32) with ESP32 module (ESP32-WROOM-32E-N4) using my custom board. My main goal is to get WiFI and A2DP Bluetooth Classic working. I am facing an issue with the performance however. I have narrowed down the (probable) cause to ESP-Hosted slave (ESP32 module) itself, (only when running ESP-Hosted-MCU slave compiled firmware).

Steps to reproduce:

• Prepare ESP-Hosted-MCU Slave example via idf.py create-project-from-example "espressif/esp_hosted=2.6.0:slave"
• Enable via menuconfig CLI Shell console and disable Bluetooth to test only wifi (simultaneously freeing memory for an iperf server)
• Compile and flash the module
• Execute commands via CLI:

wifi init
wifi start
sta SSID PASSWD
iperf -s

• Run iperf2 on PC and test performance

I am getting only about 2Mbps of UDP/TCP throughput without Bluetooth and ~700kbps when Bluetooth is enabled.

Notes:

I am using ESP-IDF version 5.5.1 and ESP-Hosted-MCU version 2.6.0

During diagnostics I have tried some other things:

• Verifying SDIO (4 bit, 20 MHz) transport speed between ESP32P4 and ESP32: Got ~40Mbps unidirectional
• Verifying ESP-Hosted on another EV board (Waveshare ESP32P4+ESP32C6 combo): Got ~30 Mbps TCP and ~40Mbps UDP unidirectional
• Trying running ESP-IDF iperf project example instead of the iperf integrated into the ESP-Hosted-MCU CLI on the ESP32, bypassing ESP-Hosted altogether: Got ~25 Mbps TCP and ~33Mbps UDP unidirectional

Any help would be appreciated.

Thanks,
Norbert

In the file below you can find full ESP32 slave logs:
ESP32 ESP-Hosted-MCU log.log

Below I am attaching only iperf results.

[...]
coprocessor>  iperf -s
I (234551) IPERF: mode=tcp-server sip=localhost:5001, dip=0.0.0.0:5001, interval=3, time=30
I (234555) iperf: Socket created
coprocessor>  I (236865) wifi:<ba-add>idx:1 (ifx:0, 48:a9:8a:06:20:7f), tid:6, ssn:2, winSize:64
I (236887) iperf: accept: 192.168.100.114,55576

Interval       Bandwidth
 0.0- 3.0 sec  1.87 Mbits/sec
 3.0- 6.0 sec  1.85 Mbits/sec
W (245385) iperf: tcp server recv error, error code: 128, reason: Socket is not connected
I (245387) iperf: TCP Socket server is closed.
I (245389) iperf: iperf exit
 6.0- 9.0 sec  1.75 Mbits/sec
 
coprocessor>  iperf -c 192.168.100.114 -i 1
I (561993) IPERF: mode=tcp-client sip=localhost:5001, dip=192.168.100.114:5001, interval=1, time=30
I (562003) iperf: Successfully connected

Interval       Bandwidth
coprocessor>   0.0- 1.0 sec  2.62 Mbits/sec
 1.0- 2.0 sec  2.12 Mbits/sec
 2.0- 3.0 sec  2.25 Mbits/sec
 3.0- 4.0 sec  2.25 Mbits/sec
 4.0- 5.0 sec  2.12 Mbits/sec
 5.0- 6.0 sec  2.25 Mbits/sec
 6.0- 7.0 sec  2.25 Mbits/sec
 7.0- 8.0 sec  2.12 Mbits/sec
 8.0- 9.0 sec  2.88 Mbits/sec
 9.0-10.0 sec  2.12 Mbits/sec
10.0-11.0 sec  2.12 Mbits/sec
11.0-12.0 sec  2.50 Mbits/sec
12.0-13.0 sec  2.50 Mbits/sec
13.0-14.0 sec  2.00 Mbits/sec
14.0-15.0 sec  2.00 Mbits/sec
15.0-16.0 sec  2.00 Mbits/sec
16.0-17.0 sec  1.75 Mbits/sec

coprocessor>  17.0-18.0 sec  1.00 Mbits/sec
W (580903) iperf: tcp client send error, error code: 104, reason: Connection reset by peer
I (580905) iperf: TCP Socket client is closed.
I (580907) iperf: iperf exit
18.0-19.0 sec  0.75 Mbits/sec

I am attaching results from the iperf on PC as well:

$ iperf -c 192.168.100.198 -i 1
------------------------------------------------------------
Client connecting to 192.168.100.198, TCP port 5001
TCP window size: 85.0 KByte (default)
------------------------------------------------------------
[  1] local 192.168.100.114 port 55576 connected with 192.168.100.198 port 5001
[ ID] Interval       Transfer     Bandwidth
[  1] 0.0000-1.0000 sec   245 KBytes  2.00 Mbits/sec
[  1] 1.0000-2.0000 sec   180 KBytes  1.47 Mbits/sec
[  1] 2.0000-3.0000 sec   141 KBytes  1.15 Mbits/sec
[  1] 3.0000-4.0000 sec   239 KBytes  1.96 Mbits/sec
[  1] 4.0000-5.0000 sec   186 KBytes  1.52 Mbits/sec
[  1] 5.0000-6.0000 sec   205 KBytes  1.68 Mbits/sec
[  1] 6.0000-7.0000 sec   177 KBytes  1.45 Mbits/sec
[  1] 7.0000-8.0000 sec   228 KBytes  1.87 Mbits/sec
[  1] 8.0000-9.0000 sec   217 KBytes  1.77 Mbits/sec
[  1] 9.0000-10.0000 sec   177 KBytes  1.45 Mbits/sec


$ iperf -s -i 1
------------------------------------------------------------
Server listening on TCP port 5001
TCP window size:  128 KByte (default)
------------------------------------------------------------
[  1] local 192.168.100.114 port 5001 connected with 192.168.100.198 port 56643
[ ID] Interval       Transfer     Bandwidth
[  1] 0.0000-1.0000 sec   275 KBytes  2.25 Mbits/sec
[  1] 1.0000-2.0000 sec   226 KBytes  1.85 Mbits/sec
[  1] 2.0000-3.0000 sec   219 KBytes  1.80 Mbits/sec
[  1] 3.0000-4.0000 sec   219 KBytes  1.80 Mbits/sec
[  1] 4.0000-5.0000 sec   240 KBytes  1.97 Mbits/sec
[  1] 5.0000-6.0000 sec   215 KBytes  1.76 Mbits/sec
[  1] 6.0000-7.0000 sec   245 KBytes  2.00 Mbits/sec
[  1] 7.0000-8.0000 sec   215 KBytes  1.76 Mbits/sec
[  1] 8.0000-9.0000 sec   233 KBytes  1.91 Mbits/sec
[  1] 9.0000-10.0000 sec   218 KBytes  1.78 Mbits/sec

[mantriyogesh]

Hi @nkp123,

Thanks for clarifying the setup and sharing the picture.

From your explanation, it’s clear that your Wi-Fi and network setup are working fine, as the P4–C6 combination achieves baseline throughput (~30 Mbps TCP and ~40 Mbps UDP unidirectional). Although that seems slightly low (if measured on the P4 side), I’ll set that aside for now.

1. Regarding your main issue with the P4 + ESP32 (SDIO) setup — could you please share a picture showing the connections you’ve made?
2. From the logs, it appears that you’re running the iperf test directly on the coprocessor. Could you confirm if this is intentional, or if you’re expecting throughput measured from the P4 host?
3. Please attach the sdkconfig files (for both sides) and the complete flashing logs as text files (not screenshots). These often reveal configuration issues more clearly.
4. I suggest disabling Bluetooth for now and focusing on achieving stable Wi-Fi throughput first.
5. Before testing Wi-Fi throughput, consider running the [raw transport-level throughput test](https://github.com/espressif/esp-hosted-mcu/blob/main/docs/troubleshooting.md#2-raw-throughput-testing) (both RX and TX). It helps verify the transport capacity independently.

[nkp123]

Hi @mantriyogesh,
Thanks for the reply.

Answering your questions:

1. Sure, here is the schematics:

The PCB is 4 layer with one whole ground plane. Traces are length-matched, however I did not do proper impedance matching. Now that I am looking at the design I see that pull-up resistor on the CLK line is probably not needed, but removing it didn't change anything.

2. My goal is to see good throughput measured from the P4 host. I just thought that measuring it directly from the co-processor will remove some variables (SDIO transport and host configuration). I think this was wrong assumption, as you will see in point 4. From now on I will test speed exclusively via P4 host.

3. I am attaching sdkconfig files and flashing logs below:
   ESP32P4 Host - sdkconfig.txt
   ESP32P4 Host - flash.log
   ESP32 Coprocessor - sdkconfig.txt
   ESP32 Coprocessor - flash.log

4. I have disabled Bluetooth on the ESP32 coprocessor and re-done tests with the networking stack on the P4 host and got some weird results:
   PC --> ESP32:

$ iperf -c 192.168.100.198 -i 1
------------------------------------------------------------
Client connecting to 192.168.100.198, TCP port 5001
TCP window size: 85.0 KByte (default)
------------------------------------------------------------
[  1] local 192.168.100.114 port 34426 connected with 192.168.100.198 port 5001
[ ID] Interval       Transfer     Bandwidth
[  1] 0.0000-1.0000 sec  4.38 MBytes  36.7 Mbits/sec
[  1] 1.0000-2.0000 sec  3.88 MBytes  32.5 Mbits/sec
[  1] 2.0000-3.0000 sec  3.50 MBytes  29.4 Mbits/sec
[  1] 3.0000-4.0000 sec  3.25 MBytes  27.3 Mbits/sec
[  1] 4.0000-5.0000 sec  3.25 MBytes  27.3 Mbits/sec
[  1] 5.0000-6.0000 sec  4.38 MBytes  36.7 Mbits/sec
[  1] 6.0000-7.0000 sec  3.75 MBytes  31.5 Mbits/sec
[  1] 7.0000-7.5877 sec  1.88 MBytes  26.8 Mbits/sec
[  1] 0.0000-7.5877 sec  28.3 MBytes  31.2 Mbits/sec

ESP32 --> PC:

$ iperf -s -i 1
------------------------------------------------------------
Server listening on TCP port 5001
TCP window size:  128 KByte (default)
------------------------------------------------------------
[  1] local 192.168.100.114 port 5001 connected with 192.168.100.198 port 50630
[ ID] Interval       Transfer     Bandwidth
[  1] 0.0000-1.0000 sec   244 KBytes  2.00 Mbits/sec
[  1] 1.0000-2.0000 sec  2.61 MBytes  21.9 Mbits/sec
[  1] 2.0000-3.0000 sec  81.6 KBytes   668 Kbits/sec
[  1] 3.0000-4.0000 sec  2.32 MBytes  19.5 Mbits/sec
[  1] 4.0000-5.0000 sec  2.42 MBytes  20.3 Mbits/sec
[  1] 5.0000-6.0000 sec  0.000 Bytes  0.000 bits/sec
[  1] 6.0000-7.0000 sec  2.68 MBytes  22.5 Mbits/sec
[  1] 7.0000-8.0000 sec   609 KBytes  4.99 Mbits/sec
[  1] 8.0000-9.0000 sec  2.33 MBytes  19.6 Mbits/sec
[  1] 9.0000-10.0000 sec  0.000 Bytes  0.000 bits/sec
[  1] 10.0000-11.0000 sec  2.27 MBytes  19.0 Mbits/sec
[  1] 11.0000-12.0000 sec  2.31 MBytes  19.4 Mbits/sec
[  1] 12.0000-13.0000 sec  0.000 Bytes  0.000 bits/sec
[  1] 13.0000-14.0000 sec   766 KBytes  6.28 Mbits/sec
[  1] 14.0000-15.0000 sec  2.86 MBytes  24.0 Mbits/sec
[  1] 15.0000-16.0000 sec  0.000 Bytes  0.000 bits/sec
[  1] 16.0000-17.0000 sec  2.30 MBytes  19.3 Mbits/sec
[  1] 17.0000-18.0000 sec  0.000 Bytes  0.000 bits/sec
[  1] 18.0000-19.0000 sec  3.01 MBytes  25.2 Mbits/sec
[  1] 19.0000-19.4187 sec   897 KBytes  17.6 Mbits/sec
[  1] 0.0000-19.4187 sec  27.6 MBytes  11.9 Mbits/sec

5. I am attaching SDIO transport speed test results:
   ESP32P4 --> ESP32

I (31527) stats: 20-30 sec Tx:53879 Rx:0 Kbps
I (41527) stats: 30-40 sec Tx:53894 Rx:0 Kbps
I (51527) stats: 40-50 sec Tx:53875 Rx:0 Kbps
I (61527) stats: 50-60 sec Tx:53885 Rx:0 Kbps

ESP32 --> ESP32P4

I (31529) stats: 20-30 sec Tx:0 Rx:64281 Kbps
I (41529) stats: 30-40 sec Tx:0 Rx:64263 Kbps
I (51529) stats: 40-50 sec Tx:0 Rx:64282 Kbps
I (61529) stats: 50-60 sec Tx:0 Rx:64265 Kbps

Bidirectional:

I (31527) stats: 20-30 sec Tx:7215 Rx:57419 Kbps
I (41527) stats: 30-40 sec Tx:7049 Rx:57657 Kbps
I (51527) stats: 40-50 sec Tx:7101 Rx:57574 Kbps
I (61527) stats: 50-60 sec Tx:7136 Rx:57531 Kbps

[SohKamYung-Espressif]

Hello, @nkp123

For #112 (comment) , can you also test the UDP performance also?

Note that when running iperf on the P4 as a udp client, you have to leave out the usual bandwidth multiplier. For example:

onP4> iperf -c <ip addr> -i 1 -t 60 -u -b 60

This will tell the P4 to send UDP packets at 60 Mbits/s.

I have disabled Bluetooth on the ESP32 coprocessor and re-done tests with the networking stack on the P4 host and got some weird results:

One possible reason for the weird TCP result is unoptimised Wi-Fi parameters. By default, the Wi-Fi parameters in sdkconfig.defaults.esp32p4 are optimised for the ESP32-C6 as the slave co-processor.

Try adjusting the Wi-Fi and LWIP related parameters in sdkconfig.defaults.esp32p4. As a starting point, you can use the values used for the ESP32 in the standard ESP-IDF iperf example https://github.com/espressif/esp-idf/blob/master/examples/wifi/iperf/sdkconfig.defaults.esp32 .

You can also try some of the parameters listed in this ESP-Hosted performance document https://github.com/espressif/esp-hosted-mcu/blob/main/docs/performance_optimization.md

Note

replace CONFIG_ESP_WIFI_xxx with CONFIG_WIFI_RMT_xxx in the sdkconfig.defaults.esp32p4 config

Bidirectional:

I (31527) stats: 20-30 sec Tx:7215 Rx:57419 Kbps
I (41527) stats: 30-40 sec Tx:7049 Rx:57657 Kbps
I (51527) stats: 40-50 sec Tx:7101 Rx:57574 Kbps
I (61527) stats: 50-60 sec Tx:7136 Rx:57531 Kbps

Bidirectional testing is usually for testing SPI-Full Duplex transfer. For SDIO, (which is half-duplex), the numbers may be misleading.

[nkp123]

Hi @SohKamYung-Espressif

Thanks for your reply and insight about the fact that bidirectional testing is not needed in this scenario.

I will try to adjust sdkconfig and will get back with results.

Meanwhile, I have made some UDP tests using the same exact configuration as in the previous comment.

As I don't want to bloat this post too much, I will only provide logs from the receiver side (as they are more meaningfull).

ESP32P4 --> PC

Sender side:

P4> iperf -c 192.168.100.114 -i 1 -t 10 -u -b 60
Bandwidth reported: 52 Mbps +/- 0.8 Mbps

Receiver side:

PC$ iperf -s -i 1 -u
[ ID] Interval       Transfer     Bandwidth        Jitter   Lost/Total Datagrams
[  1] 0.0000-1.0000 sec  4.95 MBytes  41.5 Mbits/sec   0.153 ms 1185/4715 (25%)
[  1] 1.0000-2.0000 sec  4.99 MBytes  41.8 Mbits/sec   0.355 ms 1144/4702 (24%)
[  1] 2.0000-3.0000 sec  5.02 MBytes  42.1 Mbits/sec   0.137 ms 1065/4649 (23%)
[  1] 3.0000-4.0000 sec  5.47 MBytes  45.9 Mbits/sec   0.127 ms 866/4771 (18%)
[  1] 4.0000-5.0000 sec  5.21 MBytes  43.7 Mbits/sec   0.132 ms 995/4708 (21%)
[  1] 5.0000-6.0000 sec  4.16 MBytes  34.9 Mbits/sec   0.124 ms 1715/4684 (37%)
[  1] 6.0000-7.0000 sec  5.41 MBytes  45.4 Mbits/sec   0.159 ms 866/4725 (18%)
[  1] 7.0000-8.0000 sec  5.39 MBytes  45.2 Mbits/sec   0.157 ms 888/4730 (19%)
[  1] 8.0000-9.0000 sec  5.35 MBytes  44.8 Mbits/sec   0.166 ms 871/4684 (19%)
[  1] 9.0000-10.0000 sec  4.64 MBytes  38.9 Mbits/sec   0.161 ms 1400/4708 (30%)

-b=40 brought packet loss below 5%
-b=25 brought packet loss to near 0

PC --> ESP32P4

Sender side:

PC$ iperf -c 192.168.100.198 -i 1 -t 10 -u -b 60M
Bandwidth reported: 62.9 Mbps +/- 0

Receiver side:

P4> iperf -s -i 1 -u
Interval       Bandwidth
 0.0- 1.0 sec  57.85 Mbits/sec
 1.0- 2.0 sec  58.11 Mbits/sec
 2.0- 3.0 sec  58.62 Mbits/sec
 3.0- 4.0 sec  57.89 Mbits/sec
 4.0- 5.0 sec  58.46 Mbits/sec
 5.0- 6.0 sec  58.12 Mbits/sec
 6.0- 7.0 sec  57.96 Mbits/sec
 7.0- 8.0 sec  58.11 Mbits/sec
 8.0- 9.0 sec  58.31 Mbits/sec
 9.0-10.0 sec  57.76 Mbits/sec

No packet loss results this time, however in general this looks like the maximum speed attainable - increasing bandwidth did not yield any higher throughput.

[nkp123]

Hello again @SohKamYung-Espressif,
It looks like the default configuration of the iperf example for ESP32P4 target (sdkconfig.defaults.esp32p4) is the same as for ESP32 target (sdkconfig.defaults.esp32).

I have tried other settings based on the document you have mentioned - I got the same, unstable results while transmitting from the ESP32P4 host.
Below is an example of TCP throughput with settings recommended for ESP32-C2 coprocessor:

[  1] 0.0000-1.0000 sec  2.66 MBytes  22.3 Mbits/sec
[  1] 1.0000-2.0000 sec   859 KBytes  7.03 Mbits/sec
[  1] 2.0000-3.0000 sec  1.41 MBytes  11.8 Mbits/sec
[  1] 3.0000-4.0000 sec  2.56 MBytes  21.5 Mbits/sec
[  1] 4.0000-5.0000 sec  2.37 MBytes  19.9 Mbits/sec
[  1] 5.0000-6.0000 sec  2.62 MBytes  22.0 Mbits/sec
[  1] 6.0000-7.0000 sec   928 KBytes  7.61 Mbits/sec
[  1] 7.0000-8.0000 sec  1.16 MBytes  9.70 Mbits/sec
[  1] 8.0000-9.0000 sec  2.59 MBytes  21.7 Mbits/sec
[  1] 9.0000-10.0000 sec  1.98 MBytes  16.6 Mbits/sec
[  1] 10.0000-11.0000 sec  2.53 MBytes  21.2 Mbits/sec
[  1] 11.0000-12.0000 sec  2.39 MBytes  20.1 Mbits/sec
[  1] 12.0000-13.0000 sec  1.37 MBytes  11.5 Mbits/sec
[  1] 13.0000-14.0000 sec  1.41 KBytes  11.5 Kbits/sec
[  1] 14.0000-15.0000 sec  2.36 MBytes  19.8 Mbits/sec
[  1] 15.0000-16.0000 sec  4.22 KBytes  34.6 Kbits/sec
[  1] 16.0000-17.0000 sec  2.22 MBytes  18.6 Mbits/sec
[  1] 17.0000-18.0000 sec  2.13 MBytes  17.9 Mbits/sec
[  1] 18.0000-19.0000 sec  18.6 KBytes   152 Kbits/sec
[  1] 19.0000-20.0000 sec  20.0 KBytes   164 Kbits/sec
[  1] 20.0000-21.0000 sec  2.51 MBytes  21.1 Mbits/sec
[  1] 21.0000-22.0000 sec  1.43 MBytes  12.0 Mbits/sec
[  1] 22.0000-23.0000 sec  4.22 KBytes  34.6 Kbits/sec
[  1] 23.0000-24.0000 sec  1.76 MBytes  14.7 Mbits/sec
[  1] 24.0000-25.0000 sec  20.0 KBytes   164 Kbits/sec

[SohKamYung-Espressif]

In #112 (comment)

P4> iperf -c 192.168.100.114 -i 1 -t 10 -u -b 60
Bandwidth reported: 52 Mbps +/- 0.8 Mbps

PC$ iperf -c 192.168.100.198 -i 1 -t 10 -u -b 60M
Bandwidth reported: 62.9 Mbps +/- 0

These are close to the UDP figures I get, and show that the transport and basic Wi-Fi operations are okay.

To make TCP throughput consistent, you probably have to adjust the CONFIG_LWIP_TCP_xxx in the sdkconfig.defaults.esp32p4
file.

Maybe start with these numbers:

CONFIG_LWIP_TCP_SND_BUF_DEFAULT=16384
CONFIG_LWIP_TCP_WND_DEFAULT=16384
CONFIG_LWIP_TCP_RECVMBOX_SIZE=20
CONFIG_LWIP_TCPIP_RECVMBOX_SIZE=20

You can leave CONFIG_LWIP_UDP_RECVMBOX_SIZE as it is.

Run iperf, and confirm that the TCP throughput is stable. If yes, increase CONFIG_LWIP_TCP_SND_BUF_DEFAULT and / or CONFIG_LWIP_TCP_WND_DEFAULT by 2048. Also increase the CONFIG_LWIP_TCP_RECVMBOX_SIZE and / or CONFIG_LWIP_TCPIP_RECVMBOX_SIZE by 5 or so.

Gradually increase the figures until you get a high TCP with stable throughput. This is, unfortunately, a trial-and-error method, which I also had to go through to get the performance numbers for various ESP32 SOCs as co-processors.