csciguy8
commented
6 months ago All major restructuring is done and have posted (in the PR description) the latest Google P3DT performance numbers using Cesium for Unreal performance tests.
A snippet...
GoogleTiles.LocaleDeathValley (cold cache) - 25% load time improvement
GoogleTiles.LocaleDeathValley (warm cache) - 44% load time slow down
GoogleTiles.LocaleChrysler (cold cache) - 17% load time improvement
GoogleTiles.LocaleChrysler (warm cache) - 48% load time slow down@kring , keeping you in the loop. Oddly enough, looks like we're still seeing roughly the same improvements as originally guessed back in November, when the original issue was written, 17%-25% improvements for Google P3DT scenes in Unreal. And we're doing that without increasing the max simultaneous loads (still 20).
Unfortunately, the warm cache numbers got worse. The reason is not obvious, but there may be something we can do to make this better. Will add a check box to the TO DOs.
kring
electrum-bowie
When loading tiles, implement a new staged pipeline where network requests are dispatched separately from processing work.
This achieves a 25-32% loading time reduction in some test cases.
Closes #746. Also closes #473 as a side effect.
Performance Tests Summary (Cesium for Unreal Performance Tests)
Complete Testing Data
``` cesium-native main -------------------------------------------------------------------------- SampleLocaleDenver (27 MB) Cold cache - 2.02, 2.03, 2.08 | 2.96, 2.45, 2.09, 2.15 - 2.04 avg (best 3) Warm cache - 0.81, 0.90, 0.91 | 0.98, 0.98, 0.93, 0.94 - 0.87 avg (best 3) SampleLocaleMelbourne (89 MB) Cold cache - 3.38, 3.38, 3.45 | 3.54, 4.10, 4.26, 3.54 - 3.40 avg (best 3) Warm cache - 1.75, 1.76, 1.81 | 1.96, 2.09, 1.81, 1.83 - 1.77 avg (best 3) GoogleTiles.LocaleDeathValley (33 MB) Cold cache - 4.63, 4.75, 4.91 | 5.17, 5.08, 4.96, 4.96 - 4.76 avg (best 3) Warm cache - 1.92, 1.96, 1.98 | 5.07, 2.00, 2.03, 2.02 - 1.95 avg (best 3) GoogleTiles.LocaleChrysler (62 MB) Cold cache - 6.69, 6.84, 6.84 | 7.04, 7.09, 6.98, 6.96 - 6.79 avg (best 3) Warm cache - 3.06, 3.16, 3.21 | 3.62, 3.26, 3.25, 3.26 - 3.14 avg (best 3) this PR -------------------------------------------------------------------------- SampleLocaleDenver (27 MB) Cold cache - 2.43, 2.46, 2.52 | 2.99, 2.86, 2.77, 3.05 - 2.47 avg (best 3) Warm cache - 0.97, 1.01, 1.04 | 1.12, 1.08, 1.09, 1.05 - 1.00 avg (best 3) SampleLocaleMelbourne (89 MB) Cold cache - 3.19, 3.25, 3.27 | 4.14, 5.17, 3.31, 3.84 - 3.23 avg (best 3) Warm cache - 1.78, 1.81, 1.88 | 1.94, 2.03, 2.06, 1.95 - 1.82 avg (best 3) GoogleTiles.LocaleDeathValley (33 MB) Cold cache - 3.22, 3.24, 3.24 | 3.25, 3.32, 3.38, 3.57 - 3.23 avg (best 3) Warm cache - 2.24, 2.25, 2.26 | 2.28, 2.29, 2.37, 2.38 - 2.25 avg (best 3) GoogleTiles.LocaleChrysler (62 MB) Cold cache - 5.05, 5.16, 5.17 | 5.20, 5.20, 5.37, 5.17 - 5.12 avg (best 3) Warm cache - 3.39, 3.42, 3.47 | 3.51, 3.53, 3.60, 3.57 - 3.42 avg (best 3) ```In Depth
...
Simplified view of existing tile loads  This PR introduces this structure  The best place to start looking at the code changes would be in `Tileset:: _processWorkerThreadLoadQueue`. Previous load requests were handled immediately, but now they are now passed to `TilesetContentManager`, which transforms the requests into new data that `TileWorkManager` can work with, a generalized `TileLoadWork`. `TileWorkManager` handles the critical logic to queue / throttle / and execute `TileLoadWork` objects.TO DO