[OSM-dev] [OSM-talk] donating read-only api-mirrors

Stefan de Konink stefan at konink.de
Sun Feb 8 09:15:38 GMT 2009


Matthias Julius wrote:
> Stefan de Konink <stefan at konink.de> writes:
> 
>> Matthias Julius wrote:
>>> Why would storing of the data be such a big overhead?  Especially
>>> comparing to storing everything in (limited) memory?  I think compared
>>> to all the other processing the API is doing writing the data to disk
>>> and then reading it again would not be so significant.
>> Sadly that is very significant because it becomes a serial process, thus 
>> introduces a bottleneck.
> 
> Why a bottleneck.  I didn't think this would require too many
> recources.  Of course it will delay the response to the client, but I
> don't think this should be a huge problem.

You start to write with ~300 processes at the same time contiguously to 
the disk. What possibly could go wrong?


Stefan




More information about the dev mailing list