It’s that time of year again – PDC 2010 kicks of tomorrow! Join the Silverlight Performance Team as we take you through the high level analysis of common performance issues that apps commonly run into. I’ll be giving a live session titled “Optimizing Performance for Silverlight WP7 Apps” at 3:15pm (PDT) on day 1 (28 October 2010).
Even if you couldn’t make it here to heckle me in person, the rest of the team will be online and answering your questions (or heckling on your behalf) as you watch the live stream via the online player, which you’ll be able to find at http://player.microsoftpdc.com.
Make sure to check out the session before mine (“Things I Wish I Knew about Building WP7 Apps” by Jaime Rodriguez) and all of the rest of the Windows Phone 7track. There’s other great content at PDC that’s worth checking out, but we all know that that’s secondary!
See you there!
File this one under “Sad, but True”…
Always prefix your source paths with a “/” (full-qualified path) instead of simply using relative paths.
But they both work!?!
True, both of these will work, equally well (visually), but performance wise the relative path will do extra lookups which waste time, and can hit the SD card more than you want it to (causing further slow downs). This is something that we will fix on our side in the future – since all storage is isolated and we can assume that there is always a “/” (if there is no “..”). For now, it doesn’t hurt to get the extra performance boost by simply prefixing your paths with a “/”.
As Luke Kim, a friend from Microsoft, pointed out “/” paths are not really full qualified paths. To make things clear though, I use the term “full qualified” since we are within the confines of the .NET IsolatedStorage framework and there is no access to the rest of the system, “/” paths are as fully qualified as you get (without actual URI specifiers). Thanks for pointing this out!
This is kind of obvious – but important. Read the White Paper which was written by the Silverlight performance team (mainly Shane Guillet) and browse through the samples that come with it. In these blog posts I’ll try to distill specific items from the paper into blog format, but you can’t replace the feel of the complete paper.
Additionally check out the performance video with Shane, which runs through a lot of the samples from the document and distills specific tips and tricks to get your app running smoothly: http://channel9.msdn.com/shows/Inside+Windows+Phone/Inside-Windows-Phone-03-Optimizing-Windows-Phone-Silverlight-applications/
- Do not use the built in ProgressBar straight up, use Jeff’s template
- When you’re done with an indeterminate ProgressBar, make sure to toggle IsIndeterminate to False and Collapse the bar
- General: Always make sure to stop animations / remove animating controls when they’re no longer needed
Due to a bunch of different reasons the shipping ProgressBar control is suboptimal and will actually be UI thread bound – meaning that if your UI thread is stuck working, your ProgressBar will be stuck as well. Not a great situation for a ProgressBar, huh?
That said, we’re not leaving you high and dry. Jeff Wilcox has a great solution which changes the template for the ProgressBar to only run on the Render thread – meaning that it will continue ticking, even when you’re doing your heavy loading work on the UI thread. That said, it still comes with a gotcha – don’t forget to set IsIndeterminate to False and to Collapse the bar once your done (instead of just setting Visibility to Hidden) so that the ProgressBar doesn’t continue to tick in the background, eating up Render thread cycles.
As a general rule, highlighted especially by the ProgressBar, you should always make sure to stop animations (not pause) and remove / collapse animating controls when they’re no longer needed.
Remember: just because you can’t see a animation / control doesn’t mean that it isn’t there doing work.
I’m kicking off a series of posts about Silverlight perfofmance under Windows Phone 7 with a a kind of obvious one, but one that is important to keep in mind from the get go.
- Test your code on device as much as possible
But the Emulator is awesome?!?
True, the emulator, otherwise known as the x86 Device Emulator, or simply XDE, is awesome, but it is still not an accurate representation of a device. In fact since the XDE is usually so smooth, it’s extremely easy to fall into the trap of adding more features “because it works on the emulator”.
The emulator restricts itself to one core, adds artificial Sleep()’s and limits the amount of memory it is happy to eat up (so it won’t just chew through whatever is available), but that still isn’t enough.Chances are that even running at one core the emulator is still running faster than the device (most cores today are going to be running faster than 1GHz and chances are you are going to have less things running on that core than the device does). Throw in a desktop GPU which beat a mobile GPU handsdown and you’ve got a winning combination. If you happen to have an older machine, then the XDE will simply run like a dog – and you won’t be able to tell if your app crawls because of your code or because of your machine.
But I don’t have a device!
Common problem, especially in these trying, pre-release, times. Fear not though! Your local Microsoft office most likely has some devices and can help hook you up. Shoot them an email and let them know that you are working on an app, include a description and some screenshots from the XDE (to sweeten the deal) and they should be able to help.
It’s one of those bugs… If you’ve tried profiling Silverlight lately and you’ve run into a consistent crash in Silverlight which brings down the browser, but only on specific projects then this bug is for you.
Basically, profiling any Silverlight app (plugin or OOB) that takes advantage of Shaders will cause Silverlight (and its container) to crash. The only current workaround is to remove the shaders before profiling (possibly with an #ifdef if you are so inclined). This is slated to be fixed in an upcoming version of SL 4 (though no release dates yet).
Note: people often get scared of crashes since they can indicate a security bug – but this is not a security issue (the profiler puts us into a bad state, causing the crash).
Cross-posted from msdn.
Here’s a small tip for those of you who want to debug performance in a Windows Phone Silverlight app with the frame rate counters, but have the System Tray visible – hide it.
The counters currently show up behind the system tray (since technically the tray is a system overlay which is drawing over the surface available to your Silverlight app), so hiding the tray will show the counters.
Don’t forget: to re-enable the system tray when you’re done!
Here’s an awesome gotcha when moving from a desktop Silverlight application to a Windows Phone 7 application – make sure that your media (wmv) files are set to “Build Action” = “Content” and not “Resource”.
You’ll notice that if you do something like:
Where “somevideo.wmv” is set to “Content”, then the Windows Phone Developer Tools (ie. Visual Studio) will underline the
"Source" attribute and recommend that you set it to “Resource”. This is a hangover from the desktop and is something that I hope will go away – you can safely ignore this warning (it won’t appear in your build windows).
What’s Wrong With “Resource”?
For those that want more, here are the potential problems you can run into when setting your media to “Resource”:
- When a video file is compiled as a Resource it incurs an extra space and performance hit every time you play it, since Silverlight does extra processing to extract the video from your assembly (DLL). In the case of “Content” the file can be read directly from disk (or memory) and you’ll get instant start playback.
- Anything that makes your DLL larger is evil (from my point of view) – you want your assemblies to be small (think “quick and nimble”). Although the size doesn’t always directly affect load and memory time (there are a couple of other factors at play here) this helps eliminate one more possible bottleneck.