Network setup

[hrvojezg00]

Hi all, we are having some DR issues so wanted to check if it could be network related. We have Qnap Ts-2888x with 4 10gbe intel nic`s port trunked (8 ports) to a Cisco 10 gbe switch with total of 80 gbe, so I`m not sure if it is better to have it port trunked or make all ports unlinked? Problems occur when rendering 6k resolution images with many lightmixes etc (big files) with DR, sometimes slower nodes render faster then faster ones, sometimes a node just steps out of rendering etc. Thought I ask if someone has similar setup and got it to work correctly.

Thanks,
H

[Juraj]

You mean DR? This sounds like issue on master slave, unless they also take far too different time to start-up. The latter could indicate some issue of competing for same port to draw assets. But if the issue remains after a long time, let's say hour then the issue would be on part of Master node.
I indicate the Master node congestion by looking into DR log to see how long it took to send each .exr package. They can get into gigantic filesizes with Lightmix, hence why with Lightmix I dramatically lower the amount of megapixels being sent in each package.

[hrvojezg00]

They seem to get more stable as render goes on, but doesn`t scale as it should. For lower resolution we put 30 sec sync inter, for bigger 60 sec, both 500 000 max pixels. Tried all possible combinations, this worked best. How do you set it?
Do you have ports port trunked?

[Juraj]

I don't because I am lazy to install my bigger switch (Off-topic: I have 28 port 10GBe ready but I need to cut out holes for 120mm fans on top because replacing the stock 40mm fans causes the switch to error, but 120mm fans run...file lol), so I have only enough ports right now for single port per device.
But I toyed with the idea for long time because most devices I have dual-nic Intel cards so I'll definitely get to it by summer !

I use 60 seconds by default because otherwise I would forget to change it back. I use 6 MPX (6 000 000 ) so more than 10 times what you. This way I will get full frame transfer within 6 minutes for 8k image. Even with Lightmix it shouldn't create more than +/- 2-3GB .exr package, so 2-3 seconds to transfer the package to master. I have less than 10 slaves, so no chance of saturation in this way.

You might have to analyze the DR Log on each individual slave to see if something fishy isn't happening. Or analyze the Master traffic but that is far less forward.

[hrvojezg00]

So your NAS is connected with only one cable to the switch?

[Juraj]

Yes, but that is never a problem. On average my scenes use up to 2GB of assets in terms of textures and of course 2GB of scene file, so it takes up to 5-10 seconds at start of DR for each node to transfer the file. As far as network goes, because the textures take 10 times longer to load by the DR client but there is nothing the network can do about it.
And after that the network storage (File server in my case, I use actual full-fledged PC not NAS) doesn't do anything at all until saving the file rendered file. So bottlenecks happen on master slave if they happen.

[hrvojezg00]

Thanks Juraj, I will try raising max pixel transfer and see what happens. There are 8 of us in the office so there is alot of traffic going on, so would like to hear network setups of offices with more then 10 people who use in house render farm.

[Juraj]

Thanks Juraj, I will try raising max pixel transfer and see what happens. There are 8 of us in the office so there is alot of traffic going on, so would like to hear network setups of offices with more then 10 people who use in house render farm.

You would be surprised how few have 10GBe setups :- ).

Overall though I feel is that it's 3dsMax & Corona which are more to blame for often arcane behavior of DR.

[hrvojezg00]

I've opened alot of tickets regarding DR and never got real solution to the problem. Sometimes it works, sometimes it don't...

Thanks again!
Thanks Juraj, I will try raising max pixel transfer and see what happens. There are 8 of us in the office so there is alot of traffic going on, so would like to hear network setups of offices with more then 10 people who use in house render farm.

You would be surprised how few have 10GBe setups :- ).

Overall though I feel is that it's 3dsMax & Corona which are more to blame for often arcane behavior of DR.