Hi Itzik,
From participating the group here I have heard many complaining that the driver level is the cause to problems like you are facing. From my own experience I have noticed problems when using the RAW API. The RAW API is NOT thread safe (I am using an OS) and therefore one needs to follow the guidelines and understand how to use it. LwIP is small and light weight. I have used it with ARM9 that had only 96K RAM. Actually TCP stack was assigned less then 10K... 😊 If you have 50Mb why not run an OS and run it with threads etc... I suggest looking for a nice example with Sockets + lwipopt.h and continue from there. If you are using it to stream an unreliable connection you should probably add some protocol to overcome this instability. I do not see the reasoning for defining 8M for MEM_SIZE ? You do not need so much. 100K is more then sufficient for your testing, I think. You do not more PCB's... PCB is your control block... every block you send or receive is using one. I am defining 80 TCP PCB's with 25K RAM for MEM_SIZE and I do not have so much RAM. Check some examples and see if you can find more ideas... Hope that helped, Noam. ________________________________ From: lwip-users <[email protected]> on behalf of Itzik Levi <[email protected]> Sent: Monday, November 6, 2017 11:45 PM To: Mailing list for lwIP users Subject: Re: [lwip-users] Lwip tcp-stack reliability issue when using non-reliable network? Hi Noam, Thanks for the prompt response! The actual link isn't serial. its a non-reliable rfcomm link, which is a stream based protocol. Regarding ram available ram - up to 50MB I would say. I'm attaching my entire lwipopts.h here. https://pastebin.com/24AR5sYB [https://pastebin.com/i/facebook.png]<https://pastebin.com/24AR5sYB> lwipopts.h - Pastebin.com<https://pastebin.com/24AR5sYB> pastebin.com Regarding the socket API - I did not actually put lots of thought to it, it was trivial to implement and that's that. do you have a reason to believe the root cause is there? Although the only change between the 2 tests I ran was randomly dropping data in the physical layer output? If you think that might be it, I can rewrite the socket layer and attempt to test again. And regarding misusing the stack - I totally agree, I probably did misused it somehow, I'm new to lwip and that's why I'm here. Question is, what did I do wrong, and how to pinpoint the problem. Thanks, Itzik. On Mon, Nov 6, 2017 at 11:19 PM, Noam Weissman <[email protected]<mailto:[email protected]>> wrote: Hi Itzik, I see that you have defined: TCP_WND=(100 * TCP_MSS) TCP_SND_BUF=TCP_WND MEMP_NUM_TCP_SEG=TCP_SND_QUEUELEN If TCP_MSS is defined 536 (default) that means that your TCP_WND is 53,600 bytes ? and that is over a serial line ? How much RAM do you have?, How much RAM is defined for TCP memory? I never worked with PPP so I am a bit on the dark here. From my experience with LwIP (over 6 years) Almost always it was misusing the stack or doing something wrong. You have written that you are using the Socket API. Normally Socket API is used with different threads/task. If you do not use an OS why not use the RAW API ? BR, Noam. ________________________________ From: lwip-users <[email protected]<mailto:[email protected]>> on behalf of Itzik Levi <[email protected]<mailto:[email protected]>> Sent: Monday, November 6, 2017 7:07 PM To: [email protected]<mailto:[email protected]> Subject: [lwip-users] Lwip tcp-stack reliability issue when using non-reliable network? Hi All, Firstly, a bit of background: I'm currently evaluating lwip stack in order to determine whether it can fit my needs. The end-goal, is integrating Socket API --> TCP/IP --> PPPoS stack on both sides over a serial-like-interface(stream-based) which isn't reliable. I decided to unittest it before actually integrating the stack. Setup: (This is done symmetrically on both "client" and "server") 1. Network Interface: * The "physical layer" is currently a localhost connected tcp socket. * Using pppapi in order to create the network interface. * Its "output_cb" is the tcp connect socket(as mentioned before). * Using "pppos_input_tcpip" which is fed by the same tcp socket. 2. Forming a TCP connection over the interface: * Using lwip's bsd-like socket api. * Single socket connection. * Disabling Nagle. * Working with a non-blocking sockets. * Not writing and reading at the same time(mutex protected), but using lwip_poll in parallel to both. 3. Data validation: * Transferring the same generated data client-->server and server-->client and validating at the same time. Results: * When not introducing losses to the physical layer, the data is passed and validated successfully without any issues. * When starting to randomly introduce losses to the physical layer(basically drop some data when output_cb is being called), it seems to go well for minutes, but eventually it appears that I'm loosing data when sending. I'm sure I'm doing something wrong here, can you please help me figure this thing out? Pointers to debug such a case is also welcome! I'm attaching lwip's debug logs from both client and server, when losses are introduced to both physical layer ends. In this specific case some data that was sent from the client --> server, and the server failed to validate the data. The last received valid stream offset is 19008, the next 400 bytes were incorrect(future data), basically I got some skipped data. lwip configuration: SYS_LIGHTWEIGHT_PROT=1 NO_SYS =0 LWIP_TCPIP_CORE_LOCKING=1 LWIP_TCPIP_CORE_LOCKING_INPUT=1 TCP_WND=(100 * TCP_MSS) TCP_SND_BUF=TCP_WND MEMP_NUM_TCP_SEG=TCP_SND_QUEUELEN I used lwip's repo commit: 5d8d21fcae63c36005baf1b15e91268836dec679. (which is lwip 2.0.3 plus some..) Please tell me if any more info is required. Thanks, Itzik _______________________________________________ lwip-users mailing list [email protected]<mailto:[email protected]> https://lists.nongnu.org/mailman/listinfo/lwip-users
_______________________________________________ lwip-users mailing list [email protected] https://lists.nongnu.org/mailman/listinfo/lwip-users
