Trouble shooting NDI& Dante how to turn of QOS

[Petervk]

Hey, im trouble shooting a NDI|HX multicamara setup that is performing poorly.

Its a workstation running Windows & Vmix with hardware: 3960x 24 core, rtx2080S, 64gb 3200mhz ddr4 and a 480gb 900P Optane drive with 2 x 2tb NVME drive for storage.

The workstation is connected with the 10gb port and all the latest versions and drivers.

The camera's are 3 to 5 PTZ cameras that send 4k 30p with bitrates around 20-50mbps per feed, Poe+ is used to power them from the switch with tested cables of max 15-20M

I extensively tested and optimized both the workstation and cameras already.

So in theory i should have more bandwith that i could ever want, yet still it begins stuttering and showing signs of slowdown with additional camera's.

Nothing els runs on the network, no internet, no additional hops.

One of the things i learned with Dante digital audio is when you have enough bandwith, just turn of QOS. 

Also in every guide referenced for NDI network tuning is turning QOS off one of the first points.

But for some reason i cant turn of QOS anywhere, looked at the manual, around the internet. I just need a kill switch toggle for the QOS to see if improvements are happening, why is this not in the interface? or how can i reduce any handeling of the QOS functions.

I can only choose the trust mode, and then configure it.

Its not just about the problem, i just want to be able to turn of QOS and test and see.

Hope someone can assist me with this.

[NeilS]

Hi @Petervk 

The MS510TXPP is not on the recommended list to manage NDI traffic. 

Please note that only the M4300 switch family were certified by Newtek.

Due to this, the NETGEAR Design Engineering Services team have only deployed the M4300 to support with NDI cameras and TriCaster systems, with success.

NDI is heavy mDNS and there are often conflicts with IGMP. The M4300 models were enhanced for seamless Bonjour/mDNS handling with IGMP simultaneously managing other streams.

A team member would be happy to arrange a call with you to discuss further.

If so, please send a private message at your earliest convenience, so we are able to assist further.

Many thanks...

[nmarques]

Are you using multicast or unicast?

I would also, according to NewTek guidelines, make sure Flow Control is enabled and Jumbo Frames are disabled. 

NDI|HX is also low bandwidth so you shouldn't need the 10G port. Try it with a 1G port and watch bandwidth in Task Manager, as well as CPU usage. 

What software are you seeing the poor video in? vMix? NDI Studio Monitor?

If your QoS is DSCP strict priority, I don't think it would affect NDI traffic since the packets aren't marked. 

FWIW, I have about 8x full bandwidth NDI sources running very nicely on a M4200, but I'm doing all multicast. 

[Petervk]

Hey sorry for the somewhat late response.

Im only using unicast, all camera's are directly going in one network hop to the workstation only.

Jumbo frames are disabled and i experimented with flow control but unfortunatly it does not seem to impact the situation a lot.

Also tried 1g port, no difference, CPU usage is almost nothing.

The issue also remains when running at 720p low bandwith so its really strange.

In the mean time i did get a firmware patch because i complained 20mbps was way to low for 4k30p because lots of compression artifacts where visible on the footage.

They upped the limit to 51200kbps per camera so thats nice, no real impact on the somewhat inconsistent performance of the total system outside of better quality footage ofcourse.

Also checked its also not a single core that is experiencing any real load, disabled SMT but not much difference.

Disabled any virtualisation, also no effect.

degrading video is mostly seen in vmix but that logs it the most accurate.

Did lots of vmix tuning, ofcourse large input buffers but also with the advanced settings.

But OBS and wirecast experience similar issues.

I'm not even sure its the network thats at fault, its more of a how can i optimize everything as much as possible, because such a setup should be well in the posibility of running without hickups.

Wat are you're software solutions? and did you do any big network or software tweaks for the system?

I feel it does have something to do with inconsistent delivery of frames or frame times that give dropped frames, and i can reasonly accept that if a dropped frame is a I frame it will cause a notable hickup in the image, but this is just a thought without any real foundation