Re: [lwip-users] Bad TCP throughput, delayed send frames

[Uwe Roth]

Hi Leon,

thanks a lot for your very fast help. I tried your suggestion and it works
excellent.

Regarding the actual version of lwip I have to admit that I tried to
upgrade, but it still doesn't work. My 1.1 package answers arp requests.
Thats all it currently does. I don't know why.

Your fine workaround/advice takes a lot of pressure from me and gives me
some time to look for the causes of this failure.

So, being ungratefully, may I ask you another question?

In order to achieve the desired throughput it will be necessary to send
large blocks of data and splitting the data into the largest possible tcp
frames.

You know the C16x architecture and its restrictions, for example the 16K
barrier in large memory models. This results in some limitations for the
size of memory pools and pbufs. As an example, a memory pool as realized in
lwip, I suppose, will never be greater than 16K for a C16x. There may be
other restrictions implied by the internal memory and the architecture of
the CS8900.

You know much better than I do the optimal relationship between all of the
defineable setup parameters of lwip. Do you have any hints for desirable or
possible maximum values for such values as TCP_MSS, TCP_SND_BUF,
PBUF_POOL_BUFSIZE and so on? RAM ist not a problem, there will be about 500K
available.

Thanks in advance.

Uwe

----- Original Message -----
From: "Leon Woestenberg" <address@hidden>
To: "Mailing list for lwIP users" <address@hidden>
Sent: Monday, February 07, 2005 8:19 PM
Subject: Re: [lwip-users] Bad TCP throughput, delayed send frames


> Uwe,
>
> Uwe Roth wrote:
>
> > My problem's that I need a throughput of at least 50 KByte/s, but the
> > TCP Handshake runs very, very slow and depends on how often I'm calling
> > tcp_tmr().
>  > ...
> > What am I doing wrong? Do you have any ideas? If this is 'normal', what
> > can I do to increase the throughput without changing the tcp_tmr
intervall?
> >
> I remember there being a thread on this, during the post-0.7.2 changes
> in the TCP code.
>
> 1st, upgrading to CVS HEAD should fix TCP performance.
>
> On 0.7.1 you could also try calling tcp_output() (if I am right?) after
> tcp_write() to trigger the TCP stack.
>
> Leon.
>
>
>
> _______________________________________________
> lwip-users mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/lwip-users
>