Note: that the workstation(where you initiate the render process) will always be one step ahead because the scene is already loaded in the memory.
To Optimize the Distributed/Swarm rendering, a few things could be done:
- It is always a good practice to export heavy geometry objects to V-Ray Proxy (or .vrscene) files. Proxy geometry is not written to the scene file which makes the scene size smaller and saves render time and network transfer time. This is also true for the local render.
- Copy assets locally on each Render Slave. This can help tremendously, especially for complex scenes with lots of assets of large size since the network transfer will be minimal.
- Network Speed has a significant impact on Distributed Rendering. The faster the network, the quicker V-Ray will transfer its data packages, which helps reduce render times.
- Using the Bucket Sampler with lots of Render Servers. The Progressive Image Sampler has many advantages over the Bucket Sampler, but when it comes to Distributed Rendering, the Bucket Sampler provides superior scaling and it is the recommended sampler type for DR.
- Avoid the Last Bucket stuck syndrome. When rendering animations V-Ray cannot start rendering the next frame before all the buckets are complete for the current one. This may slow down the whole process if there is an area of the image that requires a lot more sampling than the rest of it. As a result, one or more buckets can get “stuck” in a specific area, while all other machines are idle and waiting for those few buckets to complete. Lowering the bucket size can help in this case. But avoid using extremely low values, because this can have the opposite effect.
- Depending on the complexity of the scene it is possible that the first few machines to join the rendering will complete the whole image before the rest of the machines can even join. Also, if there is a difference in the amount of RAM installed on the Render Servers, it is theoretically possible that some machines with less RAM will fail to render very complex scenes.
Note: that the workstation(where you initiate the render process) is good to be a more powerful machine, so it will handle better with the Distributed/Swarn render process. Eventually, the whole render process depends on its performance.