Re:Show HN: Tilemaker – DIY vector tiles from OpenStreetMap ...
by Doctor_Fegg on Sunday July 11, 2021 @06:03AM
(#27799667)
It might be doable with 256GB. I've tried with my 144GB machine and it's too slow to be feasible. But ultimately I think 128GB will be achievable... I've got a few ideas that could potentially reduce memory usage.
For whole-planet scale, the traditional approach is to load the OSM data into a Postgres database, and then serve vector tiles from there.
Re:Show HN: Tilemaker – DIY vector tiles from OpenStreetMap ...
by Cyykratahk on Sunday July 11, 2021 @05:55AM
(#27799637)
The submitter of this issue [1] reported that 64GB of RAM was insufficient to load an 18GB PBF. Considering the planet PBF is 58GB, you're going to need a lot of RAM (and time).
I tried the example tilemaker config (4 layers, zoom 11-14) on the 519MB Australia PBF. My 24 thread PC took a little under 20 minutes to finish generating 773,576 tiles, using about 10GB of RAM.
The GitHub readme lists "You want the entire planet" as a reason not to use tilemaker. Why is that? Presumably it's excessive RAM/CPU usage during pbf conversion, or when serving tiles from the mbtiles sqlite file.
But how excessive are we talking? How big a machine would be needed to process a planet file? What tools work better with huge input files?