Hi, > I think people have already touched on most of the things I want to > mention, so I'll be brief. > - An "off switch" in tilesAtHome.conf would probably be good, for > those who are serious about unlimited maximum complexity or have set > their complexity for reasons other than CPU speed. I have introduced that into the unstable branch. A value of 0 is respected to mean "unlimited" and is not changed by the code.
> - Using the complexity of the tileset when the system value for > MaxTilesetComplexity is 0 results in very low maximum complexity > values when the first tileset is not very complex. I just made a > change in my local copy to use the ($tilecomplexity * 900 / $deltaT) > value as a first guess if MaxTilesetComplexity is 0 and the tile has a > non-zero complexity. I would be interested in that patch. > - 1% new value, 99% old value in the decaying average might not change > fast enough if the initial value (whether a user's guess, the actual > first tileset complexity, or some other value) is too high or too low. > However, this is one of those points that could probably be debated > endlessly. I did log 1000 rendering requests with both size and time and ran simulations to pick the optimal weight for the EWMA. 99% and 95% work both decently well. I also tried other schemes like add 10000 if the rendering was fast enough, multiply with 3/4 if it was to slow, but EWMA was superior to the alternatives. Patrick "Petschge" Kilian _______________________________________________ Tilesathome mailing list [email protected] http://lists.openstreetmap.org/listinfo/tilesathome
