Wait, you're saying you just go to the cloud for long-distance routes? But that's no longer routing on a mobile device! Assuming your devices have laptop-like power is a bit of a cheat too, but I guess >50% of people probably have 1GHz+ by now... as for me I have to support a fleet of older devices with as little as 400MHz single-core and 48 MB RAM, and they are multifunction devices so I can't use even half of that memory... so all my posts have come from that perspective.
If you ask me, if it hogs too much memory and CPU for a Raspberry Pi, it's not a real mobile solution. Just my opinion tho :) Processor speeds are stagnant, and slower devices will remain important as long as there is demand for low prices (so, as long as the third world exists). Non-cloud solutions will be important in some markets for the same reason. (and I always encourage people in the open-source community to think about the low end, since low-income people are major beneficiaries of OSS.) I'm talking about real life routing: (A) In 2007 a mobile device typically had only 64MB RAM and a 300MHz processor. So, if you didn't have a routing hierarchy, it would take a minute or two to compute the average route a commuter would take. But now you have many phones and tablets with 1GB RAM and a 1GHz processor, so A* can buffer all the static data in RAM and only takes a few seconds. (B) Furthermore, most devices are now Internet enabled and long routes can be calculated in the cloud using A*. I have demonstrated on many occasions that the longest routes in OpenStreetMap can be calculated fairly quickly on reasonable hardware e.g. only 30 seconds on a Xeon processor with 16 GB RAM. If a normal driver calculates all his routes with A*, the computational cost will be less than one hundredth of a cent per kilometre. Insignificant compared to the cost of fuel, man hours, mobile bandwidth etc.
_______________________________________________ Routing mailing list Routing@openstreetmap.org http://lists.openstreetmap.org/listinfo/routing
