Ethernet IP performance and cRIO

[clendon.gibson@canrig.com]

Hi folks,

I have a control system running on a cRIO 9014. We are using LabVIEW 2009 SP1. This control system must read and write from a PLC at a high frequency. (Once every 10 ms would be ideal, but we can tolerate 20 or 30ms.)

We used to do this with a Profibus module in the chassis attached to the cRIO. We would like to switch to an Ethernet/IP solution. (The PLC is an Allan Bradely Control Logix.)

We noticed a significant slow down in performance when we changed from Profibus to E/IP. So I ran the Profiler on it.

The E/IP driver calls are each taking more then 10 ms on average.

This seems like a long time to send a E/IP message and get back a response.

What kind of time frame should I expect?

[Kira_T]

Hi Clendon,

What version of Ethernet IP are you using?  Which model of the Allen Bradely Control Logix are you using?  

Thanks,

Kira T

[clendon.gibson@canrig.com]

Hi Kira,

The E/IP version is 1.0.4.3. So it is the NI Labs version. We have the commercial version comming, but the media is not here yet.

I am not sure what you are looking for on the Control Logix side. The processor is a 5573. The E/IP card is a 1756-EN2T. Is that what you meant?

[Ben-N]

Hi Clendon,

I sent you an email just a few minutes ago regarding this issue, but I'll hit the high points here.

The NI Labs version of the driver was not optimized to the same extent as the new driver, and read and write times have decreased with the updgrades.  With the newest revision of the driver you should be able to read a string tag on average around once every 12ms (give or take) depending on your controller, if the network you are on is a private network, and what the PLC you are reading from is.

With that said, the first read is going to occur much slower than following reads because the first communication has to set up resorce allocation and other such things.

Let us know if you have any more problems.

[S1ack]

Using the Industrial Communications add on for Labview (2011) - so not NI Labs (dunno what then even is).

Implicit Communications:

On ControlLogix side:

I have a generic ethernet module representing the cRIO 9074.

A cyclic task at 1ms that increments a watchdog sent to the cRIO with the fast consumed data, which is 16 bytes of data total in my application.

RPI for the generic ethernet module is set to 20ms.

On the cRIO side:

I have a timed loop set at 10ms (2x RPI) which unpacks the consumed data, and packs some produced data (also 16 bytes total).

I reflect the incoming watchdog back to the ControlLogix, and compute the count difference between the current watchdog value and I see it varying from ~35 to ~80 ms round trip.

cRIO CPU is doing lots of other things at is at ~70% Total Load.

Good enough for my application, but I suppose I could increase priority of the timed loop - CPU Load permitting.

This timed loop reports 1ms iteration duration.

Private 'cubical net' between cRIO port 2, and ENB2TXT module. But that port is sending lots and lots of data to an HMI and an inverter, and a power quality meter (ModbusTCP). CPU is Logix5561

Perhaps that helps you understand EthernetIP performance using the 'upgraded' function blocks?

[clendon.gibson@canrig.com]

Hi Ben,

This is confirming what we have been measuring with the profiler when using the UCMM tag read and tag write functions. This would appear to not be the correct solution for our product.

Thanks!

[clendon.gibson@canrig.com]

Hi S1ack,

The solution we started with was using the Read Tag and Write Tag functions. These use the slower and higher overhead UCMM protocol.

We need to switch to the class 1 assemblies as in the project you describe. I have started work on this and already have good results. I have a strange hiccup every 6 seconds which I am guessing is a periodic resource contention somewhere, but otherwise the perfomance is much improved.