Cycles distributed GPU rendering

With the Blender Pigeon 2.79 build and the cycles server (on each render node) it is possible to accomplish distributed GPU rendering on a network (just starting "cycles_server.exe --device CUDA" (option --help for more infos)).

The network render settings in the Pigeon Build could be more sophisticated (eg. finding render nodes automatically instead of typing in the IP adress).

It is really easy to use. No open instances of Blender needed on the render nodes. Just starting the cycles_server on each render node, typing in the IP adresses of the nodes in the submitting Blender instance and that's it.

It would be a good idea to integrate this functionality into Blender 2.8., since distributed GPU rendering speeds up the workflow tremendously. I wonder why this feature exists in the code of Blender 2.79 but never was activated. Are there some problems or limitation with this feature? It might be buggy somehow, but this is distributed rendering how it should be.

With this solution, you see also at every time, each - from the network - rendered tile.

The solution which is propageted at the moment is crowdrender, which has more sophisticated network settings but seems to have disadvantages compared to the cycles_server solution, eg. no realtime preview of the rendered tiles is available.