On 02/11/2010 12:52 PM, Peter Körner wrote:
> River Tarnell schrieb:
>> Peter Körner:
>>> It's more important to think about where the diff-import should be
>>> running from. I'd suggest trying it from willow but keep thinking about
>>> maybe shifting it to ptolemy..
>>
>> Does importing one minute's diffs really take so much memory than 1GB isn't
>> enough?  From what I remember of osm2pgsql, it took at least several minutes
>> to use that much memory.

My guess would be that for diffs, it won't need that much memory, as the 
cache won't be effective during imports anyway. The cache stores the 
lat/lon values of nodes in a blocked-sparse array at the id-th entry in 
the table. When building ways and relations, Osm2pgsql will either query 
the each node in O(1) from the cache or require a SQL query over the 
network. It is heavily optimized for initial import and on diffs, the 
cache will mostly not be populated anyway. Therefore I guess the memory 
doesn't matter in this case, but the extra latency of the network vs 
local process might. On the other hand on a local lan, that will 
presumably be less than a single disk seek, so not really an issue either.

>
> No i don't think it will. As I said I'd try it from willow but I can
> imagine, that eg. expiring tiles won't work this easy when osm2pgsql
> does not run on the same machine as mod_tile.

Expiring tiles is a different matter, but osm2pgsql doesn't expire the 
tiles directly so shouldn't effect where it has to run. Instead it spits 
out a textual list of tiles that should be considered expired, that then 
needs to be handled separately. The actual expiry script will most 
likely need to be run on the same machine as the (meta)tiles stored 
though, as it does require to touch and stat quite a lot of files. 
Especially with the 200 different language styles. But if the meta tile 
directory is writable from the login servers, we can experiment with 
what setup works best first before moving it over to ptolemy.

Kai

>
> Peter
>
> _______________________________________________
> Maps-l mailing list
> [email protected]
> https://lists.wikimedia.org/mailman/listinfo/maps-l


_______________________________________________
Maps-l mailing list
[email protected]
https://lists.wikimedia.org/mailman/listinfo/maps-l

Reply via email to