ASK THE EXPERTS
Vision Systems Questions
Hi, we are working with a sort of complex system with 8 GigE Allied Vision cameras running in 12Bit PixelFormat mode connected to a Switch with a 10Gig uplink to a virtual machine (ESXi) on which our software grabs the images. All Cameras are only used to take single frames, no video. At least 4 cameras are triggered by the same hardware trigger to grab a frame of an object. In the worst case even 8 Cameras may grab at the same time. In the last time we were facing problems with incomplete frames. We only have one try to grab the frame since the object will moving away. We tried with different parameters like PacketSize of 9K and Receive Buffers in the hardware adapters and virtual network adapters but this won't fix the problem for us. The switch itself does not report any errors and the network traffic seems not to be at any limit. It seems like the VM is having some trouble here. Using Wireshark to take a look at incoming GVSP Packets and outgoing GVCP Packets indicates that Packets are missing on the receiving side and requested for retransmission but only up to some limit. We can see that for ~200ms not a single GVSP Packet is received, like a short 'hickup'. What we want is to get this as robust as possible since we have up to 10 seconds before the next object triggers a new grab. So we have plenty of time to wait for the frames if only possible by the SDK. From my understanding several GVSP Parameters exist capable to be tuned that way but these are either not implemented in the SDK or some internal timeouts make it impossible. Has anyone experience with such a system and the GVSP protocol? Sidenote: We have set the camera bandwidths down to 10 Mb/s but this seems not to solve the problem. Looks like the problems lays in triggering multiple cameras at exactly the same time. Best regards
- = Company Profile
- = Answered Questions
You can look at our Streampix software and our Pulse DSA box. We can easily solve this solution for your project. Let me know if you have further questions I can help with.
Mr. Schardt, I think this needs some local debugging as you have many bottle-necks in yoursystem which are NOT related to any GEV related operations. The good thing is that the Genicam crowd is next week quite near to your office and there is a plugfest Friday 11. November in Lüttich where some people might be interested in debugging your circuit. So if you are willing to take it to Lüttich, I am will to try to bring you in the room. Regards, Werner Feith
Although I'm not familiar with the AVT drivers, many GigE drivers have parameters intended to solve this problem. Specifically, camera parameters include Frame Transmission Delay, Inter-Packet Delay, and Bandwidth Reserve. Set the Frame Transmission Delay to different values for each camera so they don't all send at once. Assuming you don't have those parameters available... It isn't clear to me whether the problem is in the switch or the VM. If the switch, you might think about a model having a chip dedicated to each port. If the VM, well, I wish you luck.