Re: [lwip-users] tcp_output does not flush

[Luke Trowbridge]

Thank you so much for the reply!  Yet, I suppose I am still somewhat 
convinced that I have setup something incorrectly.  The inefficiency of such 
an algorithm (Nagle's) seems mind-boggling.  Take the following example, 
assuming TCP_MSS = 500:

1. Client requests data from host (expecting 700 bytes)
2. Host enqueues 700 bytes, but responds with 500 bytes (because of Nagle's
algorithm?)
3. Host waits for acknowledgment from client
4. Client sends delayed acknowledgment (~200ms later) from socket layer
5. Host sends remaining 200 bytes

I have attached a Wireshark capture of this.
Is it really Nagle's algorithm causing the delay seen at step 4?
If so, it seems that such a delay will occur at the end of every transfer 
whose length is not an even multiple of TCP_MSS.

Luke


--------------------------------------------------
From: "Kieran Mansley" <address@hidden>
Sent: Tuesday, May 04, 2010 9:56 AM
To: "Mailing list for lwIP users" <address@hidden>
Subject: Re: [lwip-users] tcp_output does not flush

> On Tue, 2010-05-04 at 09:45 -0400, Luke Trowbridge wrote:
> > I suspect I am misunderstanding how I can force a flush of all
> > enqueued data
>
> With TCP fundamentally you can't.  You can disable Nagle's algorithm if
> you wish (by setting the NODELAY flag on the pcb->flags) but it's a good
> idea to read up on this before doing so so you understand the trade-offs
> involved.
>
> Kieran
>
>
>
> _______________________________________________
> lwip-users mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/lwip-users