Re: [ECOS] tcp/ip performance on ecos

[Grant Edwards <grante at visi dot com>]

On Mon, Mar 11, 2002 at 07:07:53PM -0700, Gary Thomas wrote:
> On Mon, 2002-03-11 at 18:44, Yong wrote:
> > 
> > can anyone tell me how much cpu resource(mips) is needed to run tcpip stack
> > on ecos nonstop ? I am planning to make my board a video stream server on
> > 100M lan enviroment ,but concerned about overall performance of the system,
> > my board is based on hitachi sh-3 cpu which is about 60 mips .
> >
> 
> The networking performance depends on many factors, not only the CPU
> and it's speed.  Probably the most important is the actual networking
> adapter itself - some require much more overhead than others do.

With a little profiling, it may be possible to improve things
dramatically.  I re-wrote the IP checksum routine in ARM
assembly and reduced the network stack CPU usage by almost 50%.

Like Gary said, there can be a huge difference from one
Ethernet controller to another.  Worst case: no DMA and
Ethernet CRCs in software (don't laugh, I've had to do it).

OTOH, if you've got an Ethernet controller that can do
scatter/gather DMA and IP checksums in hardware, the thing is
going to fly.

-- 
Grant Edwards
grante@visi.com

-- 
Before posting, please read the FAQ: http://sources.redhat.com/fom/ecos
and search the list archive: http://sources.redhat.com/ml/ecos-discuss