I've contributed to OSM on and off for years but I'm getting back in to it due to work and Google's new pricing being prohibitive. I've set up a server to handle tiles, geocoding and routing for the planet which is great but one pattern is each tool has to do lots of mucking around to put the nodes, ways and relations in to a reasonable order. The pbf format has the capability to be indexed but its never used. Is there a reason for that? Numerical id order makes sense for doing the dump, but not at all for any kind of processing. I'd imagine a simple geohash grouping of everything done once would improve import speeds and slicing speeds immensely since relevant nodes for a way would be physically located closely thus maximising cache hits. Is there a reason this isn't done that I've missed? If not I'm tempted to make a indexer and make the relevant patches to the tools that could benefit from it the most. asked 15 Oct '18, 06:15 cheater512 |
This help site is designed for questions and answers but not really suited for starting large discussions. It sounds like you have an interesting issue to discuss. Try posting it at https://lists.openstreetmap.org/pipermail/dev/ instead.