What is the fastest (performance-wise) way to serve PostGIS features?

Question

I'm interested in the fastest way to serve PostGIS features. I can find several WMS improvements and alternatives (MapProxy, TileCache, Mapnik...) but not so many WFS Open Source improvements and alternatives. I'm not interested in tiles, but in polygons over my base layer. Served fast.

Could you please me point in the right direction?

---

Update:

I forgot to mention that my app shows cadastral information. And cadastral polygons must be selectable (not editable, they are only highlighted when you click on them).

I'm using right now a mix of WMS + Django generating GeoJSON objects when the user clicks on a part of the map, to avoid situations like thousands of polygons at same time in the client (OpenLayers).

I don't know if I'm in the right way (should I avoid potentially hard situations like thousands of pols in the client by rendering them via WMS or any other solution? Should I improve the way the polygons are served?)

Answer 1

What about simplifying geometries?

http://www.bostongis.com/postgis_simplify.snippet

https://stackoverflow.com/questions/1849928/how-to-intelligently-degrade-or-smooth-gis-data-simplifying-polygons

Or maybe a combination of WFS, WFS with simplified features and WMS?

Answer 2

GeoServer provides a user setting to control the number of decimal places in the geometries which was put in there to speed up WFS serving.

Also make sure that the server is compressing the output only if your client can efficiently uncompress it.

Finally I've heard good things about tiny OWS in terms of speed.

Answer 3

WFS caching might suit your needs. As far as I know, there are two open source solutions providing WFS caching: TileStache and GeoWebCache.

In TileStache's home page, features section it is said "Generates vector tiles from OGR datasources in GeoJSON usable in Polymaps"

GeoWebCache has some WFS caching capabilities since version 1.1.0 (the current one is 1.2.6):

GeoWebCache 1.1.0 (finally) released

...

1.c Basic WFS caching GeoWebCache 1.1.0 supports very basic WFS caching. The results from queries are stored to disk, there are currently no functions for inspecting them or programmatically removing them. The algorithm relies on computed hashes and does not check for uniqueness, so there is a minute chance for collisions. The motivation for this development was the SHP-ZIP (zipped shapefile) outputformat in GeoServer. These files can be huge and tremendously expensive to compute. This WFS service is, in its current state, not intended to be used for more complicated cases.

http://old.nabble.com/GeoWebCache-1.1.0-(finally)-released-td22870524.html

Unfortunately I have never worked with TileStache neither GeoWebCache, so I cannot share any experience about WFS caching.

Answer 4

I'm developing a big web-gis, I work with a lot of data (about 700.000 features per layer) The only solution is the one that you are doing, WMS for rendering the features and WFS for getting feature's info. Whta I'm doing (with good performance) is:

• Geoserver (with native JAI library)
• When the user click on the map, a wms getfeatureinfo is used to get the information (both shape and data) of the selected feature (here you should implement some filtering fucntion, if you have overlapped features)
• The result of getFeatureInfo is added as feature on another layer to simulate the "selection highlight"

Answer 5

Solid State DRAM Server

http://en.wikipedia.org/wiki/Solid-state_drive#DRAM-based

"DRAM bit demand coming from the server sector will climb 93.7% on year in the second half of 2011"

source: http://www.digitimes.com/Reports/Report.asp?datePublish=2011/08/16&pages=PD&seq=206