Optimizing EtherCAT Performance when using Scan Engine

[sethmurphy]

Hello everyone,

This week I have been researching and learning about the limitations of the LabVIEW's Scan Engine. Our system is an EtherCAT (cRIO 9074) with two slaves (NI 9144). We have four 9235s, two 9237s and two 9239 modules per chassis. That means we have a total of 144 channels. I have read that a conservative estimation for each scan is 10 usec. This means with our set up, assuming we only scan it would take 1.44 msec which would yield roughly a rate of 694 Hz. I know that when using a shared variable, the biggest bottle neck is transmitting the data. For instance, if you scan at 100 Hz, it'll be difficult to transmit that quickly, so it's best to send packets of scans (which you can see in my code).

With all of that said, I'm having difficulties scanning any faster than 125 Hz without railing out my CPU. I can record at 125 Hz at 96% of the CPU usage but if I go down to 100 Hz, I'm at 80%. I noticed that the biggest factor of performance comes when I change my top loop's, the scan loop, period. Scanning every period is much much more demanding than every other. I have also adjusted the scan period on the EtherCAT's preferences and I have the same performance issues. I have also tried varying the transmission frequency(bottom loop), and this doesn't affect the performance at all.

Basically, I have a few questions:

1. What frequency can I reasonably expect to obtain from the EtherCAT system using the Scan Engine with 144 channels?

2. What percent of the CPU should be used when running a program (just because it can do 100%, I know you shouldn't go for the max. Is 80% appropriate? Is 90% too high?)

3.Could you look through my code and see if I have any huge issues? Does my transmission loop need to be a timed structure? I know that it's not as important to transmit as it is to scan, so if the queue doesn't get sent, it's not a big deal. This is my first time dealing with a real time system, so I wouldn't be surprised if that was the case.

I have looked through almost every guide I could find on using the scan engine and programming the cRIO (that's how I learned the importance of synchronizing the timing to the scan engine and other useful facts) and haven't really found a definitive answer. I would appreciate any help on this subject.

P.S. I attached my scan/transmit loop, the host program and the VI where I get all of the shared variables (I use the same one three times to prevent 144 shared variables from being on the screen at the same time).

Thanks,

Seth

[DirkW]

Hi,

It looks like you are using a 9074 chassis and two 9144 chassis, all three full with Modules and you are trying to read all the IO channels in one scan?

First of all, if you set your scan engine speed for the controller (9074), then you have to synchronize your Timed Loop to the Scan Engine and not to use a different timebase as you do in your scan VI.

Second the best performance can be achieved with I/O variables, not shared variables and you should make sure to  not allocate memory in your timed Loop.  memory will be allocated if an input of a variable is not connected, like the error cluster for example or if you create Arrays from scratch as you do in your scan VI.

If you resolve all these Issues you can time the code inside your Loop to see how long it really takes and adjust your scan time accordingly.  The 9074 does not have that much power so you should not expect us timing. 500 Hz is probably a good estimate for the max. Performance for 144 channels and depending on how much time the additional microstrain calculation takes.

The ECAT driver brings examples which show how to program these kinds of Apps. Another way of avoiding the variables would be the programmatic approach using the variable API.

DirkW