Hi,
I am not sure I understand you.
You need buffering basically when you send data from UART to TCP. The reasons are:
Not to send small chunks of data in every TCP packet and secondly to read data from
UART. Your UART does not have a FIFO and does not have any means to block or control
the rate. If you block you loose data. So for that reason you must have a way to collect data
while some of it is sent via TCP.
I can suggest using the OS queue. On the UART side you write to end of queue in the ISR.
From the TCP side you read data and send it to the remote side.
Regarding TCP to UART you do not need a buffer unless you are not able to send out
all you get for some reason. TCP menages the rate and as soon as it frees the RX buffer
it will inform the remote side about it. This way you control the rate.
My bridging software was written for RAW API and handled 6 UART's. Every UART had two
queues. One for RX one for TX.
The only module I wrote using Socket API was a Websocket client. I do not have lots
of experience with Socket API.
Hope that helped.
BR,
Noam.
From: lwip-users <address@hidden> on behalf of Erkan Ersoy <address@hidden>
Sent: Sunday, September 18, 2016 7:45 PM
To: Mailing list for lwIP users
Subject: Re: [lwip-users] long time tcp receive lwip packets lost
yes second part of mail was about uart-> tcp. Still I am not using buffer but lwip packs and sends it for me
my mss setting is
#define TCP_MSS 700
I am using socket API (I needed simultaneous sending and receiving but there is not enough documentation about non-blocking netconn)
My current confusion is about tcp->uart:
the bufferless configuration works fine at first but after a couple of hours It doesn't. It receives data as 80 byte small chunks (whole data is 2096 bytes small chunks are random parts in them) and the
data piles in somewhere in lwip. after I stop sending data, I see it continues send data through uart then it stops. So i think it emptied its buffer but if I begin to send again behaviour is the same; until I restart device. And stats show no memory is overloaded.
if I add buffer the problem may disappear for couple of hours up time but I think it will pile up data slower and become a problem in several days or something.
On Sunday, September 18, 2016 7:27 PM, Noam Weissman <address@hidden> wrote:
Hi,
1.
You must add buffering
2.
RS232 is character based but TCP is packet based.
As a result of the different data handling if you do not buffer data it means that for every byte you send from
RS232 you send a TCP packet that has one data byte. That means you have a huge overhead on every byte.
This is very un-efficient and causes the system to get very slow.
I have implemented a Bridge, used a 255 byte buffer and set two parameters: size and time.
Size: If you reach a predefined number of bytes you send them in one TCP packet.
Time: If you do not get any new data into the TCP buffer after a predefined time you send what you have.
This way you collect data from UART and send it in chunks/packets.
For example you define 100 bytes as threshold or 30ms as time threshold. Every byte added from UART
To the send buffer is counted. Once the counter reaches 100 or if no new byte is added after 30ms the
TCP buffer is sent and parameters are cleared.
I also suggest using a small MSS size (I use 536 bytes).
BR,
Noam.
From: lwip-users [mailto:address@hidden
On Behalf Of Erkan Ersoy
Sent: Saturday, September 17, 2016 1:42 PM
To: Mailing List for LwIP Users
Subject: [lwip-users] long time tcp receive lwip packets lost
I am making a serial tcp bridge device. it works in server mode I have freeRTOS 9, lwip 1.4.1 and STM32F107 device
I wrote the bridging part and for testing purposes I didn't implement buffer to TCP to serial. So Some while receiving data some data piles up in somewhere in tcp thread (I think).
I wrote a python script that sends 2096 characters every second (some lorem ipsum text). serial speed is 115200. So there is enough time to send data and free memory.
But after couple of hours (it is random. sometimes 2 hours sometimes 4 and yesterday it was 12 hours) device cant receive whole data somewhat it is 80 byte parts.and it piles up in
tcp part. And after I stop my script I continue to get data from my serial window a few seconds. After memory is empty if I start my script again still same problem.
I enabled stats to see if I am getting out of memory but it looks fine. I couldn't find where my problem is.
There is a web server and an udp server running in device (I didnt send request both server while testing)
my stats before running test (it sends stat to seperate uart every 10 seconds)
accept_function: newpcb->state: ESTABLISHED
And after the problem (Nothing seems suspicious except cachehit I could't figure out what it is)