Eric, do you know if this is working as expected? I was surprised to this large amount of resource memory usage in the browser until I realized it was the tile memory from all renderer processes.

I think this might be a dupe of 571026 ? 

Looks similar. In this case it's memory usage reported twice, once in renderer and once in browser process. I'll leave it to Eric to dup this if he thinks that's appropriate.

Yup, this is the same issue. The problem here is that we have two levels of aliasing going on here (aliasing from renderer/browser to GPU client ID, and aliasing from GPU client ID to GPU server ID). The second level of aliasing isn't currently handled, so we end up double counting the memory.

It would be really great to get this issue fixed - I think we need changes to the memory-infra code, but I'm happy to help out if there are any dump provider changes needed.

Yup we'll deal with this this Q. Due to memory UMA (See https://docs.google.com/document/d/1BQSdQ3GvH_n1xJPv_WYLDCa0h-EvL4gn4ppLvglS3Yg/edit?usp=sharing) we're going to rework the computation of all the edges an such and move it to chrome. That's going to be the right time to fix it definitely.

Adding +hjd both here and to Issue 571026 in case he has any ideas if this can be easily fixed in the existing TraceViewer code (Issue 571026 has more context).