-
Notifications
You must be signed in to change notification settings - Fork 532
IDE Performance Results
In a similar vein to Build Performance Results this page aims to measure the performance of Xamarin.Android across different versions of Visual Studio. The goal is to get a higher-level picture than what we can achieve from profiling in MSBuild.
Using an internal Microsoft tool, we are able to get timing information out of Visual Studio. This timing gives a better overall picture of the devloop, as well as the exact experience developers get inside the IDE.
In each timing, I am merely pushing the "play" button (F5). We have timings for:
-
Preparation Time
: Build + Deploy, the time it took for the Xamarin.Android MSBuild targets -
Launch Time
: app started + Mono debugging session started -
Total Time
: Preparation Time + Launch Time
From the table, what are first/second/third?
-
First
: start a debugging session after an initial checkout, after nuget restore. -
Second
: stop, new debugging session with no changes. -
Third
: stop, modify a XAML file, start debugging again.
I used the same projects from Build Performance Results with a physical Google Pixel 3 device connected via USB.
This is a comparison between:
- Visual Studio 15.9.5, Xamarin.Android 9.1.5
- Visual Studio 16 Preview 2, Xamarin.Android 9.1.103.6
Session | VS Version | Preparation Time | Launch Time | Total Time |
---|---|---|---|---|
First | 15.9 | 00:57.1 | 00:02.4 | 00:59.5 |
Second (no changes) | 15.9 | 00:04.9 | 00:01.7 | 00:06.6 |
Third (xaml change) | 15.9 | 00:05.8 | 00:01.7 | 00:07.6 |
First | 16.0 P2 | 00:31.0 | 00:03.3 | 00:34.3 |
Second (no changes) | 16.0 P2 | 00:03.9 | 00:02.5 | 00:06.4 |
Third (xaml change) | 16.0 P2 | 00:04.7 | 00:02.6 | 00:07.3 |
Session | VS Version | Preparation Time | Launch Time | Total Time |
---|---|---|---|---|
First | 15.9 | 02:17.7 | 00:02.6 | 02:20.3 |
Second (no changes) | 15.9 | 00:06.1 | 00:01.8 | 00:07.9 |
Third (xaml change) | 15.9 | 00:10.4 | 00:02.0 | 00:12.4 |
First | 16.0 P2 | 01:02.8 | 00:03.2 | 01:06.0 |
Second (no changes) | 16.0 P2 | 00:06.5 | 00:02.8 | 00:09.3 |
Third (xaml change) | 16.0 P2 | 00:08.6 | 00:02.7 | 00:11.2 |
Session | VS Version | Preparation Time | Launch Time | Total Time |
---|---|---|---|---|
First | 15.9 | 03:12.5 | 00:09.6 | 03:22.0 |
Second (no changes) | 15.9 | 00:32.8 | 00:02.4 | 00:35.2 |
Third (xaml change) | 15.9 | 00:33.9 | 00:02.4 | 00:36.3 |
First | 16.0 P2 | 02:05.4 | 00:03.4 | 02:08.9 |
Second (no changes) | 16.0 P2 | 00:27.4 | 00:02.5 | 00:29.9 |
Third (xaml change) | 16.0 P2 | 00:32.5 | 00:02.7 | 00:35.2 |
The results seem to support what we are seeing when profiling MSBuild
command-line, such as Build Performance Results. There is
an element of the timings being better than command-line, since
MSBuild.exe
is reused and other build-related performance caching
you get in an IDE, but not command-line.
We think there might be a ~1 second performance regression in Launch Time
in 16.0 P2. @jonathanpeppers will follow up with the IDE team to
figure out what is happening there.
- APK Tests on the Hyper V Emulator
- Design Time Build System
- Profile MSBuild Tasks
- Diagnose Fast Deployment Issues
- Preview layout XML files with Android Studio
- Documentation