Re: [lwip-users] lwIP malloc fail

[Navcon - Pedro Paulo Santos]

Kieran,

Thanks for the valuable informations!

I ran my program and looked for the ACKs. I observed that in the correct 
part of my program (the HTTP server) all packages are alwayes acked, but 
when I begin to transfer the large file the unacked packages appear and the 
system run out of memory. The problem is: it seems that the ACK never comes. 
It's not about performance, the transfer doesn't work, even if I wait for 
minutes.

Using what you said I concluded that the lack of memory may occur because 
the microcontroller doesn't recognize the ACK that comes (and it really 
comes, because I can see it with Wireshark), and because of that, lwIP 
doesn't free segment memory and after some tranfers it doesn't find memory 
to alloc.

I'm going to look for the reason that it can't recognize the ACK.

If anyone have a suggestion I will accept it gladly.

About the NODELAY, I already use it.

I agree I won't have a high performance with low memory, but I also think 
that even with this issue, I have to be able to transfer data of large sizes 
(larger than the microcontroller internal memory), even if it takes long 
time.

Thanks,
Pedro Paulo.

----- Original Message ----- 
From: "Kieran Mansley" <address@hidden>
To: "Navcon - Pedro Paulo Santos" <address@hidden>; "Mailing 
list for lwIP users" <address@hidden>
Sent: Wednesday, April 07, 2010 12:58 PM
Subject: Re: [lwip-users] lwIP malloc fail


On 6 Apr 2010, at 16:38, Navcon - Pedro Paulo Santos wrote:

> Hi all!
>
> I'm having a problem in trying to make a system with lwIP that sends a 
> large file with TCP from an external serial Flash memory.
>
> I use the following logic: read certain amont of data from the external 
> memory, send over TCP with tcp_write (), read more data, send again, and 
> there goes on. I made a function to handle lack of memory because of 
> Nagle's algoritm (because I have little memory and can't prepare a large 
> TCP message), and use tcp_output () to force the send of the amont of 
> data. The problem is that after some transfers (no more than 15 tcp 
> transfers) the program doesn't find memory to alloc in the following lines 
> of tcp_enqueue () function:

TCP won't free the segments until it has finished with them, which means 
waiting for an ACK from the other end.  This could take some time.

You should be able to disable Nagle's algorithm, and so avoid having to work 
around it, by setting the NODELAY flag on the PCB.

If you can address the memory in your flash directly then the most efficient 
way to deal with this would be to use a  PBUF_ROM type pbuf, which will just 
reference the flash directly and avoiding having to copy the data into a 
separate buffer.

If that's not an option then you will need to do as you are and wait for 
buffers to be freed when ACKs arrive.  To get ACKs more quickly I would 
suggest using a small TCP_MSS value.  If you don't have much memory you're 
never going to get high performance, so the tradeoff in this case is 
probably worth it.

Kieran

_______________________________________________
lwip-users mailing list
address@hidden
http://lists.nongnu.org/mailman/listinfo/lwip-users


__________ Informação do ESET NOD32 Antivirus, versão da vacina 4628 
(20091122) __________

A mensagem foi verificada pelo  ESET NOD32 Antivirus.

http://www.eset.com




__________ Informação do ESET NOD32 Antivirus, versão da vacina 4628 (20091122) 
__________

A mensagem foi verificada pelo  ESET NOD32 Antivirus.

http://www.eset.com