On Fri, Jun 06, 2008 at 07:24:20PM +0200, Frederik Ramm wrote:
> Hi,
> 
> > The advantage is that writing articles to a storage means only appending
> > or linear overwriting of the containers storage and expiring the old
> > locations. Thing about a log structured filesystems.
> 
> I don't understand this argument. You argue that filesystems cannot
> be trusted to do efficiently what they were designed to do, and that
> we would be better of coding something of our on on top?

Yep - most filesystems are written to be multipurpose. To have a mix of
large and small files as in your homedirectory. Having thousands or
millions of small files in one or thousands of directory
is not a topic which filesystems get optimized for.

> It won't be long and you'll tell us to return to the design we had
> two years ago where the PNG data was contained in blobs in the
> database, on the grounds that the database is more efficient in 
> retrieving these blobs from its one huge storage file than the 
> filesystem can serve tiles ;-)

A database is optimized to a very different end. Typically the data you
retrieve is small and the working set you are workin on is huge read -
larger than your main memory. So the database optimizes the amount of
data to read/touch by having clever indexes and query planner. Just
retrieving a predefined chunk of data and using the database only for
its index and data layout is adding overhead.

Flo
-- 
Florian Lohoff                  [EMAIL PROTECTED]             +49-171-2280134
        Those who would give up a little freedom to get a little 
          security shall soon have neither - Benjamin Franklin

Attachment: signature.asc
Description: Digital signature

_______________________________________________
Tilesathome mailing list
[email protected]
http://lists.openstreetmap.org/cgi-bin/mailman/listinfo/tilesathome

Reply via email to