Re: [lwip-users] httpd slow response

[Rick Culver]

Rejean,
Thank you for this input, that solved the problem.  I had not seen your 
previous post in the archives but may have not searched on the right topic 
(when did you post it).   Based on your input I changed TCP_MSS to 1024 and 
TCP_SND_BUF to 4*TCP_MSS and it works great.  The files transfer very fast 
now, practically no delay between the packets.  I may have to increase 
PCB_POOL_SIZE but I can test that later.

I greatly appreciate the response that everyone contributed to helping me 
run this down.
Rick

----- Original Message ----- 
From: "Rejean Groleau" <address@hidden>
To: <address@hidden>
Sent: Friday, April 24, 2009 12:03 PM
Subject: Re: [lwip-users] httpd slow response


> If your HTTP server is only sending every 200ms,
> it is probably because your client browser is waiting
> for at least 4380 bytes before ACKnowledging
> (this value is based on your 1500 bytes MSS).
>
> Since your send buffer is only 3000 bytes long,
> you will never send those 4380 byte at once,
> hence always waiting for the delayed ACK,
> coming only after 200ms.
>
> Like I said in a previous post:
>
>
> Rejean Groleau wrote:
> > I had the same problem on my platform (a Luminary Stellaris
> > microcontroller with 64KB of RAM).
> > To speed things up (from a typical 200ms to a less-than-10ms delay), I 
> > had
> > to do the following:
> >
> > 1. Use a TCP_SND_BUF value four times greater than the TCP_MSS (according
> > to RFC 3390).
> > 2. Always fill completely the TCP_SND_BUF (unless no more data is to be
> > sent).
> > 3. Set a callback through tcp_poll() to attempt resending data (after an
> > arbitrary amount of time).
> > 4. In my case (an HTTP server), use the HTTP 1.1 specification (to reuse
> > connections).
> >
> > I did this according to RFC 3390 (available at
> > http://tools.ietf.org/html/rfc3390), which states:
> >
> > If (MSS <= 1095 bytes)
> >     then win <= 4 * MSS;
> > If (1095 bytes < MSS < 2190 bytes)
> >     then win <= 4380;
> > If (2190 bytes <= MSS)
> >     then win <= 2 * MSS;
> >
> > I interpreted (read between the lines) that when this "upper-bound" limit
> > is reached, an ACK is sent back.
> >
> > Note that although my MSS is 1472 bytes, I still had to go under the 
> > first
> > condition
> > (4 * MSS instead or 4380 bytes). It seems the web browsers (or windows 
> > XP)
> > do not exactly follow this scheme.
> --
> View this message in context: 
> http://www.nabble.com/httpd-slow-response-tp23204551p23220815.html
> Sent from the lwip-users mailing list archive at Nabble.com.
>
>
>
> _______________________________________________
> lwip-users mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/lwip-users