Hello, 1) I had a long chat with James Mike about his project. we basically have a same idea of a "osm-input&output, tiles-from-stored-osm-output" machine image as to the outer interfaces. a) James Mike wants osm input&output in git (sftp access) or later some very light webservice (not the rails port, a custom c++ thing) accepting osm read/write (eg. from JOSM) b) he wants a hot rendering (in c++ with a modified mapnik build he's started work on) of tiles outdated from new osm data. Those tiles would be served over git or some http daemon
Below, I'm presenting my project's aims which are less demanding in terms of required speed CPU power (James Mike told me that his solution could possibly run in embedded devices like a mobile phone !!). What I'll write replies namely to Emilie's comment on my 2nd e-mail in this mail discussion. > Well, one quick point: database performance are going to be bad. Someone is > currently trying to do something similar in one of the mailing list. Cloud > vendor are very good for serving content like a web page. They are extremely > poor for database performance. The IO is usually catastrophic. > I wouldn't expect any real good performance out of such a system especially > for rendering. I think that using cloud technology is for the moment at > least a bit pointless. > > Emilie Laffray > For my project (indoor&outdoor mapping of small scale objects (you're able to see the coffee machine and tables rendered as rectangles or point POIs) rather small concentrations (eg. spots the size of a building, european-size (wink to north america :-) ) university campus, or a small city), I just want a working machine image that's ready to redeploy and takes : a) josm data as input over http (the rails port fits me for this) b) a sync between the rails port's db and mapnik's db. So the standard implementation we don't mind using for now is : mapnik(+postgis) rendering tiles on demand through mod_tiles,renderd,minutely or hourly change interval osmosis technique We don't mind: 1) having perfectly optimized instance images (eg. we'll have <5 buildings per instance) 2) working on the cloud despite low database IO, as the cloud is advantageous for us in that we can deploy new instances remotely with simple scripting (eg. 1 company dedicated building indoor mapping space = 1 new instance for them, or if we have 5 buildings on each of our instances already and are requested to host 2 new building's maps, we run another instance filling it with those 2...). If my project (Geopard) goes into a start up something, having a dedicated machine instead of an EC2 instance for indoor&outdoor high precision maps hosting could be part of some business model. I would appreciate having either a ready to use : - "normal" machine image (eg. vmware, virtualbox) or - Amazon Machine Images (AMI) ... with easy osm input&output + hot rendering into tiles Please give me contacts of people doing any of the two. Anyway, we will release by the end of January 2010 an AMI with a "vanilla" install of rails+timely change interval sync+mapnik, shipping with nice scripts to install (maybe reconfigure) the server with your settings (eg. for the rails port, but imagine it for mapnik and the sync part too : just edit in a single config file your wished rails db names, login+password, posgres memory limits and rails website login+password accounts, and some script you run by hand will install the rails port with those settings for you). We will try to release in this AMI, some default settings that you can reuse for having indoor mapping working (ie. tiles rendering with increased max zoom + 1 floor=1 TMS layer) possibility to disable sea limits & land zones so that your map is full white/empty at the beginning, possibility to sync timely only a bbox (not the planet file) from osm.org into your db). Our scripts will be in Ruby (regular shell ruby, not rails) I think. Who wants to collaborate on those magic deployment scripts ? In the end, maybe we could have a vanilla OSM instance be configured/installed from a php page filled by some enduser, a bit like tiledrawer does (well.. the config is in JSON data passed to the AMI at startup, but then you see installation&rendering progress in status.php). Take care, Jonathan On Fri, Jan 8, 2010 at 6:13 PM, jamesmikedup...@googlemail.com < jamesmikedup...@googlemail.com> wrote: > Ok, this looks great. > So, now, that would be my suggestion to use on this amazon webservice. > A lot of small osm files, the size of the smallest tile. > > Now, on thing did occur to me while walking the dog, we could render > multiple zoom levels at once. > Every time you render the smallest unit (tile) you would render the > sections of all the zoom levels and that should prevent you from > touching the data at the same time. > > mike > > > On Fri, Jan 8, 2010 at 6:08 PM, Dane Springmeyer <bl...@hailmail.net> > wrote: > > > > On Jan 8, 2010, at 7:19 AM, jamesmikedup...@googlemail.com wrote: > > > >> http://wiki.openstreetmap.org/wiki/Mapnik#Data_Sources > >> This webpage only talkes about posgis, is there any documentation that > >> covers the osm direct rendering? > > > > http://trac.mapnik.org/wiki/OsmPlugin > > > > > > _______________________________________________ > dev mailing list > dev@openstreetmap.org > http://lists.openstreetmap.org/listinfo/dev >
_______________________________________________ dev mailing list dev@openstreetmap.org http://lists.openstreetmap.org/listinfo/dev
