Re: [lwip-users] performance

[Sam Jansen]

Leon Woestenberg wrote:
> (I think the major part of what determines the performance on such
> configurations is the device driver and processor plus bus, not so much
> lwIP itself. (Although our TCP stack may not scale to very low network
> round-trip latencies, not sure there).

I agree with the hardware/driver issue; to me this is the biggest 
problem. As a TCP stack, lwIP doesn't seem to have any problem with 
filling up the receivers window; it does slow start correctly and so on. 
This means that given sufficient buffer size and perfect network 
conditions (100MBit network) it will do a pretty good job of filling the 
pipe.

Where lwIP doesn't shine is when the network coniditions are not so 
perfect. It's my belief that the granularity of the timers hurts it 
here, having to wait a minimum 500ms for a retransmit timeout can be 
seriously detrimental to TCP performance.

I'm unsure exactly how well it copes with a somewhat congested network; 
but as far as I can tell it lacks the sophistication of the congestion 
control mechanisms present in modern OSs today (FreeBSD/Linux/OpenBSD 
are all I have tested against).

But the original question was "does it perform well on a high speed 
network" and I'll answer that with: if it's buffer size is large enough, 
then yes. It wont get as much throughput as FreeBSD, but you'll probably 
find the difference isn't so great. Of course, this is all assuming 
perfect hardware and device drivers.

--
Sam Jansen                                           address@hidden
Wand Network Research Group             http://www.wand.net.nz/~stj2