ESP8266 late latency with continuous communication over TCP
Posted: Sat Oct 10, 2015 7:02 amWe are currently in the middle of developing continuous serial communication over WiFi using the ESP8266. We have succeeded in this so far, but we have run into a suspected problem with the module.
After several seconds of continuous operation, the communication slows down to a crawl. With response times in the beginning of 30-150ms, they quickly shoot up to 5000+ms, followed by, what looks like, a complete crash of the ESP8266 module. This is what we have logged after some time of communication over TCP: http://pastebin.com/CsBQpEgD
Using TCPView I could not find an extraneous list of connections, so why is it eventually crashing with hundreds of CONNECT FAIL messages? It very much looks like the module is running out of memory before crashing, so a memory leak is not excluded. Is there something we should do to prevent such problems? Flush something? Restart the module every now and then? Once a connection is established on the computer side, the connection is not closed. So there is a back-and-forth of TCP packets.
If we run this same software over UDP, it does not suffer from this problem. A few packet losses (rarely) aside, it does not slow down at all.
The module is connected to a WiFi router, then starting up a TCP server as well listening on UDP on the same port. This way a computer can connect to it both using UDP and TCP. The bug does not go away if I only open up a TCP server (same results, connects over ID 0 instead). I have completely ruled out the microcontroller it runs on from being the problem; it crashes during the time the microcontroller is waiting for data to come from the chip. This is when the CONNECT FAIL is eventually received.
In case this is important; this is our current routine on the microcontroller that sends back the response data to the computer using CIPSEND. Command echo is turned off. http://pastebin.com/rnUTA6wB
That said, in the beginning it works great and communication is established at ~9kbps at 115200 baud. There is no data loss. After a while, it just...dies.
Really would like to know how to resolve this, since so far I am very happy about the module's performance.
After several seconds of continuous operation, the communication slows down to a crawl. With response times in the beginning of 30-150ms, they quickly shoot up to 5000+ms, followed by, what looks like, a complete crash of the ESP8266 module. This is what we have logged after some time of communication over TCP: http://pastebin.com/CsBQpEgD
Using TCPView I could not find an extraneous list of connections, so why is it eventually crashing with hundreds of CONNECT FAIL messages? It very much looks like the module is running out of memory before crashing, so a memory leak is not excluded. Is there something we should do to prevent such problems? Flush something? Restart the module every now and then? Once a connection is established on the computer side, the connection is not closed. So there is a back-and-forth of TCP packets.
If we run this same software over UDP, it does not suffer from this problem. A few packet losses (rarely) aside, it does not slow down at all.
The module is connected to a WiFi router, then starting up a TCP server as well listening on UDP on the same port. This way a computer can connect to it both using UDP and TCP. The bug does not go away if I only open up a TCP server (same results, connects over ID 0 instead). I have completely ruled out the microcontroller it runs on from being the problem; it crashes during the time the microcontroller is waiting for data to come from the chip. This is when the CONNECT FAIL is eventually received.
In case this is important; this is our current routine on the microcontroller that sends back the response data to the computer using CIPSEND. Command echo is turned off. http://pastebin.com/rnUTA6wB
That said, in the beginning it works great and communication is established at ~9kbps at 115200 baud. There is no data loss. After a while, it just...dies.
Really would like to know how to resolve this, since so far I am very happy about the module's performance.
