[Bug] Subscriber callback appears to be limited at 100Hz with ESP32

[patrickelectric]

Describe the bug

When ping-ponging bytes around between a computer and a esp32, the subscriber side in the ESP32 appears to be limited at 100Hz.

To reproduce

Run the following code:
https://gist.github.com/patrickelectric/0bd8315894fec32cdaa69f8119de1da8

System info

ESP32-c3 and:

                   -`                 
                  .o+`                 patrick@pichau
                 `ooo/                 OS: Arch Linux 
                `+oooo:                Kernel: x86_64 Linux 6.9.7-arch1-1
               `+oooooo:               Uptime: 5d 23h 29m
               -+oooooo+:              Packages: 2949
             `/:-:++oooo+:             Shell: zsh 5.9
            `/++++/+++++++:            Resolution: 7680x2160
           `/++++++++++++++:           DE: KDE
          `/+++ooooooooooooo/`         WM: KWin
         ./ooosssso++osssssso+`        GTK Theme: 
        .oossssso-````/ossssss+`       Disk: 803G / 911G (93%)
       -osssssso.      :ssssssso.      CPU: AMD Ryzen 7 2700 Eight-Core @ 16x 3.2GHz
      :osssssss/        osssso+++.     GPU: NVIDIA GeForce RTX 4060 Ti
     /ossssssss/        +ssssooo/-     RAM: 43960MiB / 64250MiB
   `/ossssso+/:-        -:/+osssso+-  
  `+sso+:-`                 `.-/+oso: 
 `++:.                           `-/+/
 .`                                 `/

[jean-roland]

I did some latency measurements on the esp32 and for payloads <1kB latency, it hovers around 9-11ms. Seems it's pretty good for esp32 and wifi. I tried the same pico setup on PC to PC and I get 4-6ms, I get similar results by cutting pico and using ICMP pings. So the issue here seems to be wifi latency performance.

I don't know if it's applicable or not for you usecase, but now we have introduced message batching, allowing you to queue multiple data before sending a packet on the network.

The exact setup I used was PC(wifi) ->(ping) esp32(wifi) ->(pong) PC(wifi) on a wifi network as quiet as possible.