Re: [lwip-users] Max throughput and LWIP

[Siva Velusamy]

> >
> > When you say 350kbps, you mean 350 kilobytes not bits, am I right? I
> > hope I did not misunderstand. Otherwise, 2.8 Mbps must be a wrong figure.
> >
> > I'm using Atmel arm7 based Sam7x processor. It's highly probable that
> > avr and sam families use the same pepipherals so both MACs may be same.
> > I'm easly reaching 100 kBps. From some rough measurements using my
> > scope, I can say that 300 kBps is the upper limit for my case. According
> > to me there are two factors affecting my performance:
> >
>
> I didn't reply, because I thought the hardware may be too dissimilar, but
> it's also SAM7X. So in case it helps inform anything, I have also
> constructed throughput test applications on SAM7X within 64K RAM (program
> code in Flash though). It runs at 48MHz I think it was. With a carefully
> optimised configuration using the BSD sockets compatibility API I could get
> speeds up to 11 Mbit/sec on 100baseT. My driver is zero-copy (both for tx
> and rx), although using the sockets API guarantees one copy. So using the
> sequential or raw APIs would have meant even faster recorded speeds.
>

Another data point, on a very different hardware:

MicroBlaze soft processor (100 MHz): ~ 100 Mbit/s in RAW mode, ~25
Mbit/s in BSD socket mode
PPC 405 (300 MHz) - ~ 120 Mbit/s in RAW mode.

Both of these have lots of memory, but the latency to memory is on the
order of 15-20 cycles (@100 MHz).
This is with a hardware scatter gather DMA engine.

/Siva