thanks - if you look at my last post, I think those variables are set correctly now, at least the SwarmCoordinator shows me what looks like functional communication. Both of my machines are in a group called WORKGROUP, and this is specified for both AgentGrooupName and AllowedRemoteAgentGroup. Each machine also has the Other machine specified as AllowedRemoteAgentNames.
My breakthrough came after changing AgentGrooupName and AllowedRemoteAgentGroup from default to WORKGROUP, which is what my windows group is set to.
Once I get it working between two machines, I will add a third, also in group WORKGROUP, by adding its name to AllowedRemoteAgentNames after a comma.
Obviously I am no expert, but it takes up more than the 16GB RAM my local machine has, and pushes everything else, like the OS so far into swap that the machine is unusable for 20 minutes while it bakes, and the CPU usage is really low most of the time, because the data it needs is not in RAM but in swap. This kind of thing tells me it is way over taxing this machine. I even have AvoidLocalExecution turned on.
I did some major streamlining of my scene last night which seemed to have helped. I learned, to my surprise, that objects in the scene which are hidden still affect bakes. I had a lot of hidden items in my scene to make it simpler to test, but when I deleted them instead of hiding them it got way faster and uses way less RAM. And, gives better looking results. This is very counterintuitive to VFX people like me used to DCC apps which allow objects to be in the scene but not rendered.
Success at last. I set the DNS to Automatic on both machines and it started working.
It is not really solving the main problem yet, which is that when I put the settings to the quality I want, it gobbles up more RAM than I have locally, and completely kills my machine to the point that not even the mouse functions, or any other programs, so I have to wait until a build finishes (~30 minutes) to do anything like check this page or check my email.
Is there a way to limit the amount of local RAM used, and push most of the work to the remote machines?