Re: [lwip-users] lwIP performance with FreeRTOS

[Simon Goldschmidt]

Chris Ponder <address@hidden> wrote:
> Using tcpip_callback only gave 65.4 Mbps :(
> ...
>         /* Infinite loop */

I'm not sure using an infinite loop is the best example: don't you call 
tcpip_callback too often? Also, you don't check the return value...

I'd rather let your ISR check if a call to 'udpSendData()' is already queued 
and if so, only set a flag for udpSendData() to loop (maybe with a maximum loop 
count to prevent RX from stopping. On the other hand, using that technique you 
risk using 100% CPU power for sending...

>         while (1)
>         {
>                 // Store then increment the packet count
>                 *puiPacketNumber = uiPacketCount++;
>                 szPacket[4] = (unsigned char)NULL;
> 
>                 tcpip_callback(udpSendData, szPacket);
>         }
> ...
> 
> void udpSendData(void *arg)
> {
> char    *szPacket = (char*)arg;
> struct  pbuf *p;
> 
>         /* allocate pbuf from pool*/
>         p = pbuf_alloc(PBUF_TRANSPORT, ciPayloadSize, PBUF_POOL);

Pay attentention here to the fact that you now a) use PBUF_POOL instead of 
PBUF_RAM so you might end with a pbuf-chain instead of a single pbuf and...

> 
>         if (p != NULL)
>         {
>                 /* copy data to pbuf */
>                 pbuf_take(p, (char*)szPacket, ciPayloadSize);

b) pbuf_take() calls memcpy (maybe multiple times) which might slow down the 
whole thing.

Simon
-- 
Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir
belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de