How to determine PXI system bandwidth being used?

[IanDeaville(ZF)]

Hi, I have a PXI-1024 chassis with PXI-8106 controller and a selection of multi-function, counter & CAN modules.  I am considering some upgrade options, but realise that one of the limitations is the 132MB/s system bandwidth in PXI.  Is there a way i can determine how much bandwidth I am currently using, so I can see if that is causing me trouble?

The two processor cores are regularly running around 85-90% load with the RT application, and I have some timing jitter.  Would changing to the quad-core i7 PXI-8840 give me much better capability here, or do I need some more fundamental changes?

All wisdom gratefully received!

Ian

[MitchBrown · Accepted solution]

Hey Ian,

It seems there is no native way to determine the bandwidth usage of the PXI system as per this KB article:

https://knowledge.ni.com/KnowledgeArticleDetails?id=kA00Z000000kHPZSA2&l=en-GB

If you know the sample rates and resolutions of your measurements however, you could estimate the usage which would be accurate enough to see if there was a potential issue.

Moving to the PXI-8840 controller would definitely give your system more capabilities as it is a faster clocked CPU with more cores than your current controller. If you believe the bandwidth will be an issue then I would say it may be better to go for a PXIe chassis as they have a hugely increased bandwidth, into the GB/s range. This article gives more detail on the throughput of PXIe systems:

https://knowledge.ni.com/KnowledgeArticleDetails?id=kA00Z000000kHVNSA2&l=en-GB

Hopefully this helps!

Mitch

[IanDeaville(ZF)]

Mitch,

Many thanks for the information.  Following a quick calculation it seems that of the 132 MB/s I have available I am using over 160 of them. 

Hmmm...  

Nothing in the system is reporting that it isn't running as expected, but would this manifest itself simply as missed reads or slow behaviour?

I don't think I will be able to swap the entire system for PXIe, so I need to find ways to economise on my data throughput.  The most obvious approach will be to reduce sample rates where I can, and find options to trim out excess channels.  That in itself should also reduce the processor load, as it will be handling less data. 

Time to get deep into the code again...

Regards,

Ian