How to improve performance and memory usage with a large number of polygons on a canvas?

Question

The project I'm working on needs to render an ESRI shape file, which can have a large number of polygons/shapes. When I add all these polygons, lines, points, etc to the canvas I'm using, it gets really slow.

To draw the shapes on the map, I'm creating a Path object, and setting it's Data property to a StreamGeometry. I originally used a Polygon, but according to MSDN a StreamGeometry is much lighter weight.

How can I improve the performance? Would converting the finished product to a Bitmap, or VisualBrush help? Is there a more efficient way of to render all these shapes onto the canvas?

EDIT: I forgot to mention that this needs to be able to work in a Partial-Trust XBAP.

Answer 1

No need to resort to GDI, you just need to move down a layer in the WPF API's and combine your geometry into fewer visuals. Pablo Fermicola has some useful information about picking which layer to use depending on your performance needs.

I've managed to get excellent performance using DrawingVisual and the DrawingContext class.

Answer 2

You could try using GDI+ or direct x instead of WPF.

I did a similar project (rendering Map data via WPF), for an MSDN magazine article I wrote about a year ago.

It was just something I wrote rather quickly (the article + the app took about 1 week), and was designed to just be a "this is something neat you can do", and so I didn't focus on performance much. In any case, I ran into the same problem. Once the number of polygons on the canvas got large, the rendering performance started to suffer. For example, resizing the window would involve about a 1/2 second delay or so.

A lot of this has to do with the overhead of WPF. It's primarily designed as an GUI toolkit, rather than a high performance graphic engine. That means it focuses on feature richness over efficent rendering. With smaller numbers of objects (what is likely to be found in most GUI applications), the performance is pretty good and the databinding, animation, and declarative style features (and all the event routing and other things they require) can really come in handy.

When drawing a map, however, the sheer volume of data can cause all those neat data binding features to cause perf problems, as you seem well aware.

If you don't need declarative styling and animation, then I would just eliminate WPF, and use GDI+ to draw the map your self. It basicly involves setting up the right transformation matrix to get the map to draw onto a control surface, and then iterating through all the polygons and calling a bunch of DrawPolygon methods.

To enable interaction you will have to write your own hit testing code, and you will have to redraw the map anytime the form is resized. You will also have to hand code any animations or style changes or things like that you want to do (such as highlighting regions when the mouse is over them). But, it shouldn't be that difficult to write that code. I would imaging you could do it in about 1.5 weeks.

Doing it that way, however, should improve performance, because it would come down to doing only about 20-30K vector transformations, which doesn't take much processor power on most modern CPU's. You could also look into using Direct X, which would allow you to take advantage of the GPU as well, which could give an even bigger performance boost.

Answer 3

You may be limited by the performance of the canvas widget. If you have the option of using a 3rd party toolkit look at QT. It has a high performance canvas widget that is designed to render complex shapes quickly. This has already been used for at least one GIS application, so it has some track record in this space.

You can wrap QT as an ActiveX control or use the Qyoto/Kimono .Net bindings to interface to a .Net application. Troll Tech have just revamped their website and I can't find the downloadable demo they used to have there but it shows the QGraphicsView widget rendering a very large vector drawing and zooming out in real time.