[OSM-talk] Bounding box

[Raphaël Jacquot]

Nick Hill wrote:
> Hello Oliver
> 
> Replication is certainly one of the methods we'll have to use in future. 
> However, I don't see that we'll have the resources any time soon to guarantee to 
> return insane queries.
 >

said insane queries may be due in part
* because the storage driver is not the proper one for the usage we have 
(for instance, if we're using myisam)
* because the request needs to read data that is in a separate table, 
and for some reason locks said table (because of limitations in either 
the storage driver, the database engine or both)
* because the app doesn't use transaction facilities if available in the 
database engine (which in turn would allow it to enable things like 
"copy on modify" for the current transaction -- that is, the transaction 
that modifies data sees its version while it's being processed, other 
read only transaction see the previous version, other write transactions 
are blocked if they attempt to access this particular data)

> Unfortunately, the mechanism to filter insane queries (which were perhaps 
> generated through malice or ignorance) catches a few sane queries, which 
> occasionally will require modified workflow.
> 
> Once we have upgraded the RAM in the API, we can consider increasing the node 
> limit. When we partition the database or implement tiles, we can consider 
> increasing the bbox limit.

there is no need to use partitionning or other 'hackish' schemes if the 
indexing engine features the proper index generation algorithms (such as 
R-tree based indexes)
Of course, you'll then tell me that the index takes a lot of space or 
something. well, that's what computing is all about, compromises.
that is, either your data is very compressed and needs a lot of 
computing power to sift through, or, you sacrifice storage space to have 
a fast to go through data structure, and your data is fast to access. 
you can't have both.

> 
> _______________________________________________
> talk mailing list
> talk at openstreetmap.org
> http://lists.openstreetmap.org/cgi-bin/mailman/listinfo/talk
>