Armin,

thank you for your comments! I thought about SMI, too (since we had SMI issues with every new hardware generation). 
Though our typical "does a SMI steal our cpu tests" did not show any problem.

Shared interrupts may be an issue: the network card shares the interrupt with the devb-ahci and another network card. In
 case I replace the pci-express with a pci card (same devnp-e1000.so), I don't see any missing interrupts?! That's 
really strange!!

BR
Matthias

Hi Eli,

pretty much like you. We start an own io-pkt for each EtherCAT Master.

... umount / destroy interface from default io-pkt and: 

/sbin/io-pkt-v4 -i 1 -ptcpip prefix=/dev/ecat1 -P 200 timer_pulse_prio=200
mount -T io-pkt1 -opci=0,vid=0x8086,did=0x10d3,priority=221,promiscuous,speed=100,duplex=1,name=ethercat,int_mod=0 devnp
-e1000.so

-
Matthias

The only way you are going to get a unique interrupt is by using MSI/X.



On 2013/11/6, 3:47 AM, "Matthias Ecker" <community-noreply@qnx.com> wrote:

>gave it a try on the same hardware and within a special bootimage (very
>default stuff, a network stack and EtherCAT driver + Test Application).
>The image does not contain any harddisk driver and thus the networkcard
>is running without shared interrupts (with devb-ahci)
>=> result: no missing interrupts at all.
>
>By try and error, I think the harddisk driver is really "the bad guy".
>
>Anyone an Idea, how to separate the devb-ahci from the rest?
>Is it possible to reprogram the pic in order to give the devb-ahci driver
>an own interrupt (since the Advantech bios uses 0xA for nearly every
>device, leaving a lot of interrupts unassigned)?
>
>Thanks!
>Matthias
>
>
>
>_______________________________________________
>
>Networking Drivers
>http://community.qnx.com/sf/go/post106563
>To cancel your subscription to this discussion, please e-mail
>drivers-networking-unsubscribe@community.qnx.com

I'm running in similar issues with the "testpmd" example of DPDK from
Intel. It is possible to send with testpmd every 1us a small packet, but
I see also gaps of 300-400us. Attached is a tiny test program. Its test
loop runs on my 1.8GHz machine for 1.2ms ... the max jitter was 340us !

My question is: are we observing side effects of the OS or do we test
the real-time capabilities of the mother board ??

--Armin


Matthias Ecker wrote:
> gave it a try on the same hardware and within a special bootimage (very default stuff, a network stack and EtherCAT 
driver + Test Application). The image does not contain any harddisk driver and thus the networkcard is running without 
shared interrupts (with devb-ahci) 
> => result: no missing interrupts at all. 
>
> By try and error, I think the harddisk driver is really "the bad guy". 
>
> Anyone an Idea, how to separate the devb-ahci from the rest? 
> Is it possible to reprogram the pic in order to give the devb-ahci driver an own interrupt (since the Advantech bios 
uses 0xA for nearly every device, leaving a lot of interrupts unassigned)?
>
> Thanks!
> Matthias
>
>
>
> _______________________________________________
>
> Networking Drivers
> http://community.qnx.com/sf/go/post106563
> To cancel your subscription to this discussion, please e-mail drivers-networking-unsubscribe@community.qnx.com
>

I'm running in similar issues with the "testpmd" example of DPDK from
Intel. It is possible to send with testpmd every 1us a small packet, but
I see also gaps of 300-400us. Attached is a tiny test program. Its test
loop runs on my 1.8GHz machine for 1.2ms ... the max jitter was 340us !

My question is: are we observing side effects of the OS or do we test
the real-time capabilities of the mother board ??

--Armin


Matthias Ecker wrote:
> gave it a try on the same hardware and within a special bootimage (very default stuff, a network stack and EtherCAT 
driver + Test Application). The image does not contain any harddisk driver and thus the networkcard is running without 
shared interrupts (with devb-ahci) 
> => result: no missing interrupts at all. 
>
> By try and error, I think the harddisk driver is really "the bad guy". 
>
> Anyone an Idea, how to separate the devb-ahci from the rest? 
> Is it possible to reprogram the pic in order to give the devb-ahci driver an own interrupt (since the Advantech bios 
uses 0xA for nearly every device, leaving a lot of interrupts unassigned)?
>
> Thanks!
> Matthias
>
>
>
> _______________________________________________
>
> Networking Drivers
> http://community.qnx.com/sf/go/post106563
> To cancel your subscription to this discussion, please e-mail drivers-networking-unsubscribe@community.qnx.com
>