MicroPython
Portenta H7 TCP Latency Issue #13518
-
|
Hello, I ran into a weird problem with TCP Connections on the Portenta H7 and I couldn't find any mention of it online Here are some measurements that should make it clear >>> measure_rtt()
Received b'Hello, server'
Round-trip time: 511.019 ms
>>> measure_rtt()
Received b'Hello, server'
Round-trip time: 306.276 ms
>>> measure_rtt()
Received b'Hello, server'
Round-trip time: 408.367 ms
>>> measure_rtt()
Received b'Hello, server'
Round-trip time: 508.853 ms
>>> measure_rtt()
Received b'Hello, server'
Round-trip time: 512.186 ms
>>> measure_rtt()
Received b'Hello, server'
Round-trip time: 73.175 ms
>>> measure_rtt()
Received b'Hello, server'
Round-trip time: 510.502 ms
>>> measure_rtt()
Received b'Hello, server'
Round-trip time: 512.188 ms
>>> The code I used: def measure_rtt(host='address', port=21212):
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
s.connect((host, port))
start_time = time.ticks_us()
s.sendall(b'Hello, server')
data = s.recv(13)
end_time = time.ticks_us()
print('Received', repr(data))
print('Round-trip time:', time.ticks_diff(end_time, start_time) / 1000, 'ms')
s.close()Server (on a PC): def start_server(host='0.0.0.0', port=21212):
with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
s.bind((host, port))
s.listen()
while True:
conn, addr = s.accept()
with conn:
print('Connected by', addr)
while True:
data = conn.recv(1024)
if not data:
break
conn.sendall(data)Has anyone else run into this problem? |
Beta Was this translation helpful? Give feedback.
Replies: 18 comments 14 replies
-
|
Hi, Thanks. |
Beta Was this translation helpful? Give feedback.
-
|
@IHOXOHI Did you run |
Beta Was this translation helpful? Give feedback.
-
|
Hi Robert, |
Beta Was this translation helpful? Give feedback.
-
|
During the time that I have your attention, |
Beta Was this translation helpful? Give feedback.
-
|
I know almost nothing about using Smart Phone for something else than working with the built-in apps like Phone, browser, WhatsApp. I played a while ago with a Linux emulation on the smart phone, but that did not get me an access to the USB port as serial device, as far as I recall. Same with an iPad. A simple Laptop or chromebook may be better suited for that usage. |
Beta Was this translation helpful? Give feedback.
-
You may be interested in the Micro REPL app. |
Beta Was this translation helpful? Give feedback.
-
|
For sure. Thanks. |
Beta Was this translation helpful? Give feedback.
-
|
INFO: this build requires mboot to be installed first |
Beta Was this translation helpful? Give feedback.
-
|
During very little time, I have used the appi "UserLAnd" with the archlinux distri, ..., there was an acces to the dev port!!! But, just a litlle moment... Now it seems to be a "milicious" app... |
Beta Was this translation helpful? Give feedback.
-
The config files for Arduino Portenta H7 tell, that mboot is not used. b.t.w.: I used Termius for a while. Looked at UserLand as well, but did not like it. |
Beta Was this translation helpful? Give feedback.
-
|
The logs... |
Beta Was this translation helpful? Give feedback.
-
|
Maybe I look at the wrong file. Yes, it expects mboot or another bootloader. The instruction on the firmware download files do not mention it. The board config files do not mention MBOOT, but the stm32h747.ld file mentions an Arduino bootloader. So I assume it's present in the device when sold. Do you get in dfu mode when you push the blue button at power-on? |
Beta Was this translation helpful? Give feedback.
-
|
Arduino boards don't use mboot. Did you switch the board to DFU mode ? Double tap the reset button the green LED will blink, then |
Beta Was this translation helpful? Give feedback.
-
|
Yes, I think. The led change of state... The ACM0 is present. |
Beta Was this translation helpful? Give feedback.
-
|
In dfu mode /dev/ttyACM0 is NOT present. |
Beta Was this translation helpful? Give feedback.
-
|
So the board isn't. |
Beta Was this translation helpful? Give feedback.
-
|
I have pressed twice the reset button, the led has changed of color and rythme, and the ACM0 is still present. |
Beta Was this translation helpful? Give feedback.
-
|
what is the output of |
Beta Was this translation helpful? Give feedback.
-
|
The first one is after double press the reset button. |
Beta Was this translation helpful? Give feedback.
-
|
Could you try running |
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
|
So the device is in dfu mode. You should be able to download the firmware, maybe try again using |
Beta Was this translation helpful? Give feedback.
-
|
YES! |
Beta Was this translation helpful? Give feedback.
-
|
The board is impressive! |
Beta Was this translation helpful? Give feedback.
-
|
So, with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
|
Beta Was this translation helpful? Give feedback.
-
|
Hey, |
Beta Was this translation helpful? Give feedback.
-
|
Yo, |
Beta Was this translation helpful? Give feedback.
-
|
Getting back on topic here, I can reproduce the original reported latency issue of around 500ms. It occurs on any MicroPython port that uses lwIP (which includes the Portenta H7). The reason is because of this lwIP code in #ifndef TCP_TMR_INTERVAL
#define TCP_TMR_INTERVAL 250 /* The TCP timer interval in milliseconds. */
#endif /* TCP_TMR_INTERVAL */
#ifndef TCP_FAST_INTERVAL
#define TCP_FAST_INTERVAL TCP_TMR_INTERVAL /* the fine grained timeout in milliseconds */
#endif /* TCP_FAST_INTERVAL */
#ifndef TCP_SLOW_INTERVAL
#define TCP_SLOW_INTERVAL (2*TCP_TMR_INTERVAL) /* the coarse grained timeout in milliseconds */
#endif /* TCP_SLOW_INTERVAL */Data that's written to the TCP socket is not actually committed to the wire (sent as an ETH packet) until Sending lots of data through a TCP socket has relatively high bandwidth, because this 500ms timeout is only used for small, on-off writes of data. You can try rebuilding the Portenta firmware with this interval changed. In #define TCP_TMR_INTERVAL 50 |
Beta Was this translation helpful? Give feedback.
-
|
Hi thank you very much for looking into this, we ended up using udp for everything in our project but this is good to know in the future. Also maybe a bit off-topic, I found that with non-blocking udp socket.recvfrom it seems like only one message is buffered independently of size, so calling it multiple times until EAGAIN comes up only ever receives one message. Everything else is discarded. Do you know if there is also some option that could cause this? |
Beta Was this translation helpful? Give feedback.
-
|
Commit b1e9602 has basically fixed this TCP latency issue, by enabling the Nagle algorithm. You can now also do
Right... currently the lwIP bindings in MicroPython only store one outstanding incoming UDP packet. That could be improved to have a queue of packets. |
Beta Was this translation helpful? Give feedback.
-
|
Hey, |
Beta Was this translation helpful? Give feedback.
-
|
May I suggest starting a seperate discussion on that topic rather than piling on this one? |
Beta Was this translation helpful? Give feedback.
Getting back on topic here, I can reproduce the original reported latency issue of around 500ms. It occurs on any MicroPython port that uses lwIP (which includes the Portenta H7).
The reason is because of this lwIP code in
src/include/lwip/priv/tcp_priv.h: