> some of these algorithms could run on GPU clusters these days No matter where code actually runs you need to have and handle this servers.
For reference basic requirements for a render node include: 80 GB RAM (at least; better 128 GB); 6 or more CPU cores (12+ with HyperThreading, CPU year 2011 or newer eg: Xeon X5675); Storage: 1 TB SSD for database, 1.5 TB HDD (better SSD) for storing tiles; Fast network connection with high usage or unlimited traffic (at least 30 TB/month); ( from https://wiki.openstreetmap.org/wiki/Servers/Tile_CDN - requirements for donated render server) And yes, with higher budget (for example, with more donated servers) it would be a smaller problem. > the quality of OSM's cartography is held back due to limited server > infrastructure. Note that anyone may develop software for making maps with high quality labels. It could be running locally/on small areas/very long etc. I am not involved in this area, but I am not aware about anything either giving good results or reaching some fundamental problems with OSM data format. Even dealing with conflicting labels for points, automatic smart dealing with long labels and multipolygon labelling is as far as I know completely unsolved. Nov 6, 2020, 23:19 by and...@torger.se: > > Sorry for replying to myself, but I forgot to mention one important aspect > that I myself hadn't realized until recently: it's seems to be a whole lot > about processing power too. > > > Name tag scaling and placement strategies are expensive algorithms compared > to what we the default style does now, and I see repeatedly when various > improvements to openstreetmap-carto is discussed that "the idea is good and > would improve the style, but is unfortunately too computationally expensive > so it's not feasible". I suspect that least some of the naming falls into > that category, especially when doing smart things when zooming out to give > good overview maps. > > > I have not understood why there are these CPU limits, if it's "just" due to > under-financed server infrastructure, or if it is a problem that can't be > solved regardless of server infrastructure. As a layman one would think that > some of these algorithms could run on GPU clusters these days, but I have no > idea... it feels a bit problematic though if the quality of OSM's cartography > is held back due to limited server infrastructure. > > > /Anders > > > On 2020-11-06 22:51, Anders Torger wrote: > > >> >> I'd love to help out if the workload and chance of success was reasonable, >> but I'm a bit wary about the tagging proposal process. Most of my mapping >> contributions is simple things like correcting and adding roads so all the >> various online route planners (and indeed bike computers) that use OSM in >> one way or another can work in the areas I spend time. For that the map does >> not need to be complete at all, I just need a graph of roads, and I use the >> regular government-provided maps to actually scout the area. >> >> >> Recently I got more interested in trying to make actual complete and good >> cartography, make maps that accurately describes the area (to a certain >> detail level) and doesn't require "a real map" on the side for scouting, in >> other words make OSM to be a real map in the areas I live. It would also be >> nice if one could make hiking maps for the mountains. This is an extremely >> ambitious goal, in Scandinavia, and probably many more countries, we are >> used at having really great cartography from a special cartography institute >> which is a part of the government. Previously the maps were expensive to get >> and you could only get it on paper. Today the main aspects exists for free >> in digital form (which is a good thing, it's made with tax payers' money >> after all). Here, this is the gold standard for a general-purpose map. >> >> >> However, when I see there are some key features missing in OSM to be able to >> reach that level, and each of those little features may take years of >> processing from proposal to actual implementation in a renderer (and even if >> a proposal goes through, I suppose it's not guaranteed that it may be >> implemented), then it feels like it's just too much for me, as I'm involved >> in many other volunteer projects too. Mapping is not even my main project. >> >> >> To make this happen it seems like I will end up with having to implement my >> own style and have my own tile server and using my own tags... it's just not >> feasible. What I have done so far in my own mapping applications which works >> sort of fine is to use ready-made government maps in a custom layer for the >> more zoomed out map (and indeed have an own tile server for that), and then >> switch to OSM for the most zoomed in levels. The limitations in name >> handling and missing names for large areas is less noticed when fully zoomed >> in. But it would be really cool if one could use OSM for the whole >> cartography experience. >> >> >> As far as I understand, OSM is supposed to be a decentralized and >> semi-anarchistic consensus community that's why the proposal process looks >> like it does, but somehow I was hoping for that there was a strategic work >> group with access to professional cartography expertise that on their own >> could recognize, pick up, and implement both the feature and the guideline >> for baseline type of "must have" features, while tagging proposal process >> would be for more exotic things. >> >> >> I'm afraid that with this thorough long-haul process and still pretty basic >> cartography features lacking, we may never see them. I understand that OSM >> is a geo database, not a map as such, and it seems like the actual >> map-making hasn't been a top priority but left to third parties, and this >> may be a reason that features required for top quality cartography has been >> left unimplemented for so long. >> >> >> >> Another thing with these naming features is while they are indeed important >> to reach professional-grade maps, you need to be a very patient and >> persistent perfectionist to actually care (sort of like an old-school >> cartographer), and have the endurance to continue to care. It's much easier >> to just skip the names that can't be mapped, or make them as a point and not >> care that zoomed out maps don't work well. We've seen plenty of >> desperate/chaotic use of place=locality tag just to get names when there is >> no real support. >> >> If that's the case, then it maybe is better to just relax, let go, and let >> OSM be what it is today and not try achieve what it can't do. For me this >> means going back to just doing road work, and switch to the government maps >> anytime I need a real map. I'm fine with that. >> >> >> >> On 2020-11-06 20:19, Andrew Harvey wrote: >> >> >>> All great points there, I've ran into many of those myself. If you're >>> interested in helping work on solutions to these, discussion here is >>> probably the best place to start, once ideas gain some momentum you can >>> start a tagging proposal >>> >>> https://wiki.openstreetmap.org/wiki/Proposal_process>>> . Going through >>> that process takes a huge amount of time, effort and communication, but >>> usually results in well rounded documentation and considers a wide range of >>> scenarios and creates better tags than just "using whatever tags you like". >>> >>> _______________________________________________ >>> Tagging mailing list >>> Tagging@openstreetmap.org >>> https://lists.openstreetmap.org/listinfo/tagging >>> >> >> >> >> >> _______________________________________________ >> Tagging mailing list >> Tagging@openstreetmap.org >> https://lists.openstreetmap.org/listinfo/tagging >> > > > >
_______________________________________________ Tagging mailing list Tagging@openstreetmap.org https://lists.openstreetmap.org/listinfo/tagging