[etherlab-users] general performance questions

Georg Sassen gsassen at batalpha.com
Thu May 12 15:57:42 CEST 2011


Hi,

On Wed, 2011-05-11 at 17:58 +0200, martin troxler wrote:
> We use Ubuntu 10.04 with RT_PREEMT kernels 2.6.31.11-rt or 2.6.33.9-rt
> (We started with 8.04 and 2.6.24, too) 

Ah, ok, another Ubuntu and RT PREEMPT user, so we are not the only
ones ;)

> 
> Our system works on a recent Core2Duo, has an update cycle of 250µs
> and one domain with up to 1kB (5-20 servo drives running in cyclic
> position mode).
> The CPU load of the realtime process is below 15%. 

Hmm. We only have a cycle of 1 or 2 milliseconds, but if we create many 
domains, i.e. one for each drive, even if they don't exist at all on the
EtherCAT network, we get those datagram timeout errors. If we stay below
a certain number of domains, it's also ok with a load of about 10% IIRC.

> 
> 
> Do you use the generic driver or the patched ethercat e1000 driver? 

We are using the patched e1000 driver, perhaps we should have another
look at the generic one.

> 
> If you have to use the generic driver, the interrupt throttling of the
> e1000 should be disabled! 

Good to know, in case we switch to it...

Thanks

  Georg





More information about the Etherlab-users mailing list