[OSM-dev] all things to all people

Frederik Ramm frederik at remote.org
Tue Apr 22 15:39:57 BST 2008


>> I think I'd trust Tom more than anyone else in OSM to understand the
>> load which a given feature is going to place on the servers,
> Thats not the point, its not about x number of CPU seconds per request
> - its that the statistics are fundamentally different for, say,
> running a WMS and running a tile server. One scales, one doesn't.

ALL problems regarding read-access scalability can be solved by  
implementing a proper "feed" mechanism (or DB slaves or whatever you  
want to call it). We have made steps in that direction with the  
weekly dumps, daily diffs, hourly diffs, and OsmXAPI is doing its  
best to keep a current data set with these services. Tiles at Home read  
requests have already moved off the main API to OsmXAPI, and OsmXAPI  
scales perfectly. The only thing that isn't perfect about OsmXAPI is  
that its data may still be two hours old which is good enough for  
many cases, but still not "the real thing" (not good enough to base  
an export tab on).

I think if we invest a bit more work in that direction, then we'll  
end up with a central database that will only serve live editor read/ 
writes, and everything else can be run from the slaves. It's not  
exactly a low-hanging fruit but not that difficult either, and the  
reward is not having to worry about numbers of read requests.


Frederik Ramm  ##  eMail frederik at remote.org  ##  N49°00'09" E008°23'33"

More information about the dev mailing list