Why doesn't Swarm distribute the end of lighting build?

I’ve been doing some crazy lightmass building using Swarm distributed rendering but have realised that there’s a massive section at the end of the build which doesn’t get distributed. When the percentage of completion reaches the 90s the other machines drop off leaving one machine to do the rest of the work. However, this section easily takes the longest out of all the building.

Would anybody be able to tell me what calculations are happening here and why they can’t be distributed? If there’s no way to distribute this, what can I do to reduce the build time at this stage? I feel like increasing lightmap resolution has had the biggest impact so far.

Hi CaptParis,

Once your lighting build completes use the Statistics Window to see what actors were causing lighting to spend the most time on. You can follow the information here to see how to do this: A new, community-hosted Unreal Engine Wiki - Announcements - Epic Developer Community Forums

When lighting builds are distributed the processes are sent to individual cores to be processed. When something like this happens the statistics window can help you identify the cause. It could be anything from a large lightmap resolution, complex mesh, or instanced foliage that has be calculated. It just depends, but this can help you get started in tracking down the cause.

-Tim

Thanks Tim, that’s perfect!

I also just wanted to see if you had any idea about this lighting build issue I’ve been having. It’s a bad one and I’m not sure whether to report it as a bug:

No guarantees I’ll look at it in more depth, but doesn’t look like a bug. Looks like low quality lightmaps streaming in, and not to be confused with low lightmap resolution. You can test this by disabling texture streaming in the Project Settings > Rendering.

That’s great information. I’ve narrowed it down to being caused whenever I pushed lightmaps from 1024 to 2048 resolution, is there any reason doing that would break the texture streaming? Turning off streaming altogether does seem to solve the issue, but that’s probably not a good solution in the long run right? Thanks for looking into it, I understand if you don’t have more time for it at the moment.

Followed up on your other thread.