[OSM-dev] Proxy for the API

[Martijn van Oosterhout]

On 4/16/07, Frederik Ramm <frederik at remote.org> wrote:
> Not if you do all writes on the central server.
>
> We briefly talked about this at the Essen workshop (but without deciding
> on anything). It should be possible to offload almost all read requests
> from the central server by distributing the data in a sensible way. I
> believe that having one machine only handling write requests is a
> concept that will carry us for some time to come; it's the read requests
> that will explode.

Yeah, that's the conclusion I came to: a write-through proxy. If the
upstream server is down the database becomes read-only, which is still
an order of magnitude better than what we have now.

What left me with one tiny problem: if you have multiple of these
proxies and someone does an update, how does the proxy know its
information is out of date. I went so far as designing a OSM data
proxy invalidation protocol, but havn't decided if that's even a good
approach...

Have a nice day,
-- 
Martijn van Oosterhout <kleptog at gmail.com> http://svana.org/kleptog/