I’ve always thought that on the spatial support, MSSQL was way behind many of the other database servers, in its lack of supporting storage of geometry. With the new .NET CLR you can actually add your own .NET-based object types and I’ve also tried implementing the Simple Features Specification in SQL Server. There are some limitations that made me give this up though. First of all, a user data type cannot be more than 8000 bytes. That is at most no more than 500 vertices in a geometry object, which is far too little for an accurate coastline for instance. Another problem is that SQL Server doesn’t support inheritance chains, so you can’t make a good object-oriented implementation of your datatype either.

…so yesterday I went for a completely different and much simpler approach. I decided to just store the geometry as Well-Known Binary in an image column. The reason for using an image column is that it can hold up to 2Gb of data, which should be sufficient for most geometry objects ;-). A binary field has the same 8000 byte limitation as UDT so this is no good. In addition to the geometry field, I create four real-type fields, holding the min/max values of the geometry envelope. This makes it efficient to do boundingbox based queries on the data. Any additional fields would be attributes for the geometry object.

I implemented the whole thing using SharpMap. First I created a small upload application that takes a shapefile, creates a table in the database and uploads the geometry and attributes to it. SharpMap has the necessary data readers and WKB formatters for this. The second part was to create a data provider from which SharpMap could draw from. I more or less based this on the PostGreSQL/PostGIS data provider for SharpMap, by changing the boundingbox query to use the four envelope fields. All this was not much more than an hour’s work, so it is very simple to accomplish.

I must say I was very surprised by the performance of the approach. It is just slightly faster than the shapefile data provider, which used to be the fastest data provider for SharpMap. In comparison the PostGreSQL/PostGIS is generally 4-6 times slower.

I have created a small demo web-application you can download from here. It contains two pages: one for uploading to the database, and one for rendering a layer from the database. All you need to do is to add an empty SQL Server 2005 Express database to the \App_Data\ folder and name id "GeoDatabase.mdf".

Download SharpMapSqlServer.zip (181,74 KB) (updated May 20, 2006)

Update: The MsSqlProvider is now also included in v0.9RC1, including a method for uploading from any of the SharpMap datasources to MS SQL.