We are moving slowly towards Unreal for some projects.
We have a renderfarm counting more than 50 nodes
I have a couple of question regarding installing swarm on all the nodes to distribute the render process :
Is there a way to have a offline installation of unreal? Downloading 50 times 5Gb doesn’t sound like the most optimized scenario, but if not, we’ll live with it.
Do we need to install the complete unreal dev kit on the node to be able to render out through swarm?
Our farm consist mainly of xeons that don’t meet the minimum requirements. As we only intend to render out through swarm, is there a way to by pass this issue?
You just need your slaves pc to have the DOTnet folder containing the swarm files (swarm agent .exe and coordinator.exe). Don’t need the whole engine
you can find the folder in C:\Program Files\Epic Games\4.9\Engine\Binaries. Put the folder on your slaves and open the coordinator and enter your settings!
Technically swarm slaves don’t need a gpu since the calculations are made on the cpu and ram. Not sure about your question #3.
I’m not an expert but I’ve followed this very well made guide and it worked. If everything is well configured you don’t have to send the job ‘‘manually’’… When you start building the lighting it will automatically use all the nodes that the swarm agent can see on the network.
I had this problem. DotNet framework is the culprit here. Open the swarm agents and see whether Agents are able to ping to the coordinator. Do the reverse also(i.e ping from coordinator to the clients).
Download and Install latest dotNet framework from microsoft website hopefully it may solve your problem.
Show your settings tab, on main pc and at least 1 other.
PS: I Used this great tutorial for setting up swarm on 2 computers. It shows how to set up on more. I personally could only get 2 computers working at a time, but that’s likely because of my lack of understanding of DNS and IPconfig. Something i would like to ask you about after you’ve sorted your own problem
The tutorial is for UDK but is the exact same process for unreal
Follow the above tutorial. Make sure the “Settings” corresponds to your network. Enter “*” for “AllowedRemoteAgentName” or enter Name of all Agents manually. Make sure AgentGroup is same for all.
One more problem could be DNS name resolution. In my network I always face this problem. I need to enter IP of the coordinator system otherwise, clients fail to get Job from coordinator.
Hey guys,
Thanks I finally maneged to get it working on the farm.
Needed to change the defaults settings. Why wouldn’t it work with the default settings anyways?
Another question :
Once I send a build, I have few nodes that works, but the other just try to build the scene for ever :