Route GuidesRoutes City GuidesCities Map Log in

Updates again

11 Jun 2016
Find a better bike route. Try our map & route-planner »

Become a supporter

Richard,

This is going to sound ungrateful and I apologise for that.  I really think the infrequent updates of the route map are tremendously limiting to the usefulness of cycle.travel.  Many of the threads here come down to openstreetmap errors.  These can be fixed in a moment but it can be weeks before that change is available in c.t.  For example I've just made an OSM correction in Clifford Slocombe's 'Unrouteable section' thread, it could be a long time before Clifford can test that.  I've also corrected the designation of way 324043140, I now need to wait who knows how long before I can add it to route I'm planning. No doubt I will find another block on the route a couple of segments further on and have to wait again.

So what to do?  I suggest two things:

You've already told us it takes your server a long time to process the OSM dataset.  What is the spec of that machine?  Maybe we can crowdsource some upgrades or a dedicated updates machine.

I don't think on it's own that is really an end to the problem though even if you were able to run updates every night.  Would it be feasible perhaps, if you did have a dedicated machine, to add a command to the popup menu to re-evaluate the connections off a segment?  Taking the second example above I can route into way 148073615, if I could click on it and select 're-evaluate connections' as an immediate job it would find a new path into 324043140.

I hope I don't sound ungrateful, cycle.travel is a brilliant resource and it deserves to succeed.

Julian

Comments

Sun 12 Jun 2016, 21:07

Unfortunately the routing algorithm (‘contraction hierarchies’) is all-or-nothing. Since one change potentially cascades up to form part of a better route across the whole country, and all the routes are precomputed, there’s no way to update individual sections. 

cycle.travel’s bike-specific routing further complicates it by doing a whole bunch of preprocessing over and above the raw routing algorithm, which is how it integrates real traffic data and the like. Without that it’d take one day rather than three, but to be honest I’ve seen people try the raw, unimproved OSRM bike mode and it’s not too pretty – unless you like cycling across ploughed fields!

I think a lot of it is growing pains for OSM, and in particular for bike routing on OSM, but that it’s slowly reducing over time. Certainly there are many fewer connectivity problems than when cycle.travel first started out, when I had to spend a lot of time fixing broken roundabouts and one-way streets. Roads through military areas such as the one mentioned in Clifford’s post are certainly a pain point and I’ve been thinking about setting up a MapRoulette challenge to get them fixed, though that needs a bit of thought first to determine a tagging scheme for “yes, you can cycle through here, but you need to phone the MoD first”…

Sun 12 Jun 2016, 21:43

I see what you mean about a new segment cascading up but is it strictly necessary for edge cases such as mine and Clifford's?  Would is be so bad if short term a new segment could be forced into a route by use of way points even though it was still ignored by long distance routes?  There are plenty of canal towpath segments that are hard to include in a route due to the preference rules in place, they need waypoints between every node that leads away.  This led to my question about waypoint limits for example

Being able to force a path need not be the same as it being 'naturally' selected.  The latter may come after a full routing run, or in the case of a truly obscure location, it may never be on any preferred route between any conceivable two points.  So be it. Try routing the Grand Union through Milton Keynes for example, it'll take every one of a zillion opportunities to go another way.  That's fine, following a particular route despite better alternatives is another edge case and way points cater for it.

I know nowt of contraction hierarchies but will try to educate myself to avoid further stupid questions. 

Mon 8 Aug 2016, 12:10
Any idea when the next update will be Richard?  July got skipped, will there be one before Aug 19?
Tue 9 Aug 2016, 13:35

In a week or two when I'm back from holiday! :)