On 11/12/2020 09:59, Nick Whitelegg via Talk-GB wrote:
In the early stages I think we could run it on cheap hosting hardware,
like most projects in the OSM ecosystem. I suspect for a while usage
would be light and limited to those in the OSM community. I use
Hetzner for my hosting (OpenTrailView, Hikar, MapThePaths) - I pay
around EUR 19/month but that is for a larger system that has to deal
with the whole of Europe rather than just the UK.
https://www.hetzner.com/cloud?country=gb
<https://www.hetzner.com/cloud?country=gb>
The second-lowest spec of these, the CPX11 is giving you 2GB RAM and
40GB disc space for EUR 4.19 a month. OK we'd need more than that long
term, but I suspect that would get us going in the early stages.
That'll depending on what you want the server to do, I think. For an OSM
Carto Map style with automatic updates and reasonable performance you'll
probably need > 6Gb memory for the whole of the UK these days. Maybe a
CX31 at €11 per month (i.e. about the price of a couple of pints and a
"substantial" pork pie for those in tier 2)?
https://map.atownsend.org.uk is a CX41 I believe, and renders Mapnik /
Carto CSS map tiles that cover UK and Ireland. It could probably include
another "medium sized OSM country" in the same map style as well without
too many problems.
On the question of "could we show feature X" (e.g. "cycleways with
foot=yes" different to "cycleways with foot=no) the answer is
technically yes, but you need to decide which subset of features you
want to show because there simply aren't enough ways of visually
distinguishing things that users can actually tell apart, especially
when combined with other features.
As an example, have a look at the legend at
https://map.atownsend.org.uk/maps/map/map.html#zoom=16&lat=-24.98988&lon=135.10862
. That shows:
* designation (public footpath / bridleway / retricted byway / BOAT /
UCR / none)
* width - either "narrow" (not wide enough for a 4 wheeled vehicle) or
"wide" (wide enough)
* trail_visibility
* some surface information (unclassified unpaved roads rendered
differently to paved roads)
* tunnel yes/no
* long ford yes/no
* bridge yes/no
* embankment yes/no
* long distance foot / bicycle / horse riding routes
* access=destination and =private viewed from a pedestrian perspective
and of course combinations of the above.
It does not show:
* explicit OSM keys (e.g. footway/cycleway/path/bridleway)
* explicit OSM access tags (e.g. "foot=yes or no on a cycleway")
* undesignated cycleways differently from other undesignated paths
In order to one of those (for example just "displaying cycleways as
cycleways") you'd need to remove something else that's already rendered,
otherwise users won't be able to tell features apart.
Assuming that people are planning to go down the mod_tile / Mapnik /
Carto CSS route, I'd suggest:
1. decide what zoom levels you want, which will influence exactly which
software to use
2. deciding where to start from (e.g OSM's Standard style, mine, or a
different one altogether)
3. deciding exactly what you want to change
4. make those changes,
5. see what "unintended consequences" have occurred
6. fix those and iterate round until happy
Assuming you can deal a couple of hours overnight downtime while the
database reloads I'd suggest doing most of the "deciding what to show as
different things" work in lua and the "deciding what to show it as" in
Carto CSS. It's much easier to understand and to maintain.
With regard to the "boring bit" (scripts to load databases, keep
databases up to date etc.) most of the stuff used by
https://map.atownsend.org.uk is public (links to everything are at the
top of the changelog). Much of the rest (e.g. automatic https
certificate renewal) is standard and is documented in 1000s of other
places around the internet. If anyone wants any help or advice with any
of the above please just ask.
There may be a temptation to think "the end goal is a phone app , so
actually we probably want to look at $some_other_technology instead". I
would strongly suggest following a well-trodden path first while so that
the things that are new to whoever is doing this are have
well-documented solutions. I haven't yet found a vector tile stack that
is (a) well documented and (b) free of vendor lock-in that could go on
https://switch2osm.org/serving-tiles/ yet, for example. Once whoever is
doing this is familiar with things, trying something a bit more
off-the-wall will be more likely to work without everything breaking.
The biggest requirement is for someone to actually commit to doing the
work to set something up - nothing will happen without this. If OSM UK
are happy to fund a server, and for it to fit in their DNS somewhere
then that's one less expense to worry about - but someone still needs to
do the work.
Best Regards,
Andy
_______________________________________________
Talk-GB mailing list
Talk-GB@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-gb