Re: [lwip-users] lwip-users Digest, Vol 107, Issue 32

[Chen]

Simon, thanks for your reply.

is setsockopt( sock, IPPROTO_TCP, TCP_NODELAY, (char
*)&flag, sizeof(flag) ) the same as
tcp_nagle_disable(myconn->
pcb.tcp
)?

I already have
tcp_nagle_disable(myconn->
pcb.
tcp)
in my codes when I observed packets size alternating
between 200 and 1400 bytes


> Chen <address@hidden>
> wrote:
> > I use lwip_send (....0), along with FreeRTOS on AVR32, to send out
> TCP
> > packets
> > 
> > Each time I call, I present 200 bytes to lwip. From Wireshark, I can
> > see the packets size alternating between 200 and 1400 bytes with
> more 
> > idle time right before the 1400-byte packet
> > 
> > Is there a flag I can set so that lwip sends out the packet (200
> > bytes) if the output queque is empty, without trying to packaging
> the 
> > data to make it more efficient?
>
> It is efficient as it is, as it uses less header bytes per data byte when
> combining multiple 200-byte-chunks into one 1400-byte-packet. Note that
> this is only done when there is un-ACKed data - this is the TCP
> representation of your "output queue is empty". Note that when
> using sockets, the actual output queue is always empty as tcp_output() is
> called after each send().
>
> What you want is less latency, I guess (i.e. sending the packets right
> away without combining them)? This can be achieved by disabling the nagle
> algorithm, which can be done like this on most socket
> implementations:
>
>
> int flag = 1;
> ret = setsockopt( sock, IPPROTO_TCP, TCP_NODELAY, (char *)&flag,
> sizeof(flag) );
>
>
>
> Simon