Re: [lwip-users] lwIP hangs on transfers with many short packets

[address@hidden]

multiplex wrote:
> Hi all,
>
> We're having a problem that stumps us. Using lwIP1.3.0 (Xilinx Microblaze
> port), we are using lWIP to support a telnet CLI connection. As long as we
> don't have a lot to print, no problem; once we print a lot of data
> typically, this can be even packets containing 1 byte of payload, lwIP hangs
> (typically, after 30-40 packets). With larger chuncks, it just takes longer.
>
> We are using socket mode, all threads are running at the same priority.
>
> With the debugging turned on, we see that the cause appears to be inside
> tcp_enqueue which keeps going around in circles trying to
> allocate/deallocate something but hitting a boundary.
>    
Do you mean "tcp_enqueue: too long queue 89 (max 89)"? If so, just 
increase TCP_SND_QUEUELEN in your lwipopts.h. This is a limit on how 
many pbuf shall be allowed in the queue of unacknowledged segments per 
tcp pcb. Up to (and including) 1.3.2, lwIP created a new pbuf for each 
call to tcp_write. This results in queuelen being quite high when 
calling tcp_write (or the socket functions send/write) with many small 
chunks.This has been improved in CVS HEAD already.

Simon