MSSQL Serialization

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

MSSQL Serialization

Paul Ramsey
Dan found this link https://msdn.microsoft.com/en-us/library/ee301648(v=sql.105).aspx

It’s really interesting, I wish I could ask them questions :) 
Some good use of flagging to keep things like points small. We come out slightly larger, I think, because of padding to keep the coordinates double aligned.
They do not appear to bother with alignment. Maybe because they expect to always be on x86 architectures and others that don’t fault on unaligned access? Though there’s still a (theoretical) performance penalty to unaligned access even on x86.

P.

_______________________________________________
postgis-devel mailing list
[hidden email]
http://lists.osgeo.org/mailman/listinfo/postgis-devel
Reply | Threaded
Open this post in threaded view
|

Re: MSSQL Serialization

Mark Cave-Ayland
On 21/01/16 22:49, Paul Ramsey wrote:

> Dan found this
> link https://msdn.microsoft.com/en-us/library/ee301648(v=sql.105).aspx
>
> It’s really interesting, I wish I could ask them questions :)
> Some good use of flagging to keep things like points small. We come out
> slightly larger, I think, because of padding to keep the coordinates
> double aligned.
> They do not appear to bother with alignment. Maybe because they expect
> to always be on x86 architectures and others that don’t fault on
> unaligned access? Though there’s still a (theoretical) performance
> penalty to unaligned access even on x86.

Yeah that's an interesting read :)  These days the world is getting less
x86 and in just about every test I've done there is significant
performance improvement with aligned vs. unaligned accesses. For this
reason I'd expect PostGIS to perform better on high CPU, large geometry
processing but then there are plenty of other factors e.g. caching that
could also come into play here.


ATB,

Mark.

_______________________________________________
postgis-devel mailing list
[hidden email]
http://lists.osgeo.org/mailman/listinfo/postgis-devel