Skip to main content

Experimental: Reducing the size of .NET Core applications with Mono's Linker

The .NET team has built a linker to reduce the size of .NET Core applications. It is built on top of the excellent and battle-tested mono linker. The Xamarin tools also use this linker so it makes sense to try it out and perhaps use it everywhere!

"In trivial cases, the linker can reduce the size of applications by 50%. The size wins may be more favorable or more moderate for larger applications. The linker removes code in your application and dependent libraries that are not reached by any code paths. It is effectively an application-specific dead code analysis." - Using the .NET IL Linker

I recently updated a 15 year old .NET 1.1 application to cross-platform .NET Core 2.0 so I thought I'd try this experimental linker on it and see the results.

The linker is a tool one can use to only ship the minimal possible IL code and metadata that a set of programs might require to run as opposed to the full libraries. It is used by the various Xamarin products to extract only the bits of code that are needed to run an application on Android, iOS and other platforms.

I'll add this line to a nuget.config in my project's folder. Note that NuGet will inherit global settings and ADD this line.

<?xml version="1.0" encoding="utf-8"?>

<configuration>
<packageSources>
<add key="dotnet-core" value="https://dotnet.myget.org/F/dotnet-core/api/v3/index.json" />
</packageSources>
</configuration>

Then I'll add the IL Linker's NuGet package to my project with this command line command (or from Visual Studio):

dotnet add package ILLink.Tasks -v 0.1.4-preview-906439

The assemblies will automatically be "trimmed" when they are published (not built) so I'll build it twice, disabling it with a switch:

D:\github\TinyOS\OS Project>dotnet publish -c release -r win-x64 -o notlinked /p:LinkDuringPublish=false

Microsoft (R) Build Engine version 15.3 for .NET Core

TinyOSCore -> D:\github\TinyOS\OS Project\bin\Release\netcoreapp2.0\win-x64\TinyOSCore.dll
TinyOSCore -> D:\github\TinyOS\OS Project\notlinked\

D:\github\TinyOS\OS Project>dotnet publish -c release -r win-x64 -o linked
Microsoft (R) Build Engine version 15.3 for .NET Core

TinyOSCore -> D:\github\TinyOS\OS Project\bin\Release\netcoreapp2.0\win-x64\TinyOSCore.dll
TinyOSCore -> D:\github\TinyOS\OS Project\linked\

And here's the results:

image

You can also run it with  /p:ShowLinkerSizeComparison=true and get a nice table. I've trimmed the table as it's super long.

  TinyOSCore -> D:\github\TinyOS\OS Project\bin\Release\netcoreapp2.0\win-x64\TinyOSCore.dll

Before linking (B) After linking (B) Size decrease
----------- ----------- ----------- -----------
Total size of assemblies 48,025,824 16,740,056 65.14%
----------- ----------- ----------- -----------
TinyOSCore.dll 36,352 36,352 0.00%
Microsoft.Extensions.Configuration.dll 24,584 24,584 0.00%
Microsoft.Extensions.Configuration.Abstractions.dll 20,480 20,480 0.00%
Microsoft.Extensions.Configuration.Binder.dll 24,064 24,064 0.00%
Microsoft.Extensions.Configuration.FileExtensions.dll 22,528 22,528 0.00%
Microsoft.Extensions.Configuration.Json.dll 24,072 24,072 0.00%
Microsoft.Extensions.DependencyInjection.dll 46,600 46,600 0.00%
Microsoft.Extensions.DependencyInjection.Abstractions.dll 35,336 35,336 0.00%
Microsoft.Extensions.FileProviders.Abstractions.dll 17,920 17,920 0.00%
Microsoft.Extensions.FileProviders.Physical.dll 31,240 31,240 0.00%
Microsoft.Extensions.FileSystemGlobbing.dll 39,432 39,432 0.00%
Microsoft.Extensions.Options.dll 26,120 26,120 0.00%
Microsoft.Extensions.Options.ConfigurationExtensions.dll 16,904 16,904 0.00%
Microsoft.Extensions.Primitives.dll 33,800 33,800 0.00%
Newtonsoft.Json.dll 639,488 639,488 0.00%
Microsoft.CSharp.dll 1,092,096 392,192 64.09%
Microsoft.VisualBasic.dll 465,416 0 100.00%
Microsoft.Win32.Primitives.dll 18,968 4,608 75.71%
Microsoft.Win32.Registry.dll 85,008 0 100.00%
SOS.NETCore.dll 54,264 0 100.00%
System.AppContext.dll 14,336 2,560 82.14%
System.Buffers.dll 14,336 2,560 82.14%
System.Collections.Concurrent.dll 206,360 31,744 84.62%
System.Collections.Immutable.dll 2,378,264 0 100.00%
System.Collections.NonGeneric.dll 96,792 24,576 74.61%
System.Collections.Specialized.dll 88,608 15,360 82.67%
System.Collections.dll 326,664 52,224 84.01%

TinyOSCore -> D:\github\TinyOS\OS Project\bin\Release\netcoreapp2.0\win-x64\publish\

You can see in some places where there's no size decrease. That's because I'm using those assemblies completely. Some see a 100% decrease - they've been removed entirely - because I'm not using the Registry, for example. And some see a fractional decrease because I'm using some methods but not others.

You can check out the full instructions and try this yourself at https://github.com/dotnet/core/blob/master/samples/linker-instructions.md. Again, it's a work in progress.


Sponsor: Check out JetBrains Rider: a new cross-platform .NET IDE. Edit, refactor, test and debug ASP.NET, .NET Framework, .NET Core, Xamarin or Unity applications. Learn more and download a 30-day trial!



© 2017 Scott Hanselman. All rights reserved.
     


from Scott Hanselman's Blog http://feeds.hanselman.com/~/452296766/0/scotthanselman~Experimental-Reducing-the-size-of-NET-Core-applications-with-Monos-Linker.aspx

Comments

Popular posts from this blog

dotnet sdk list and dotnet sdk latest

Can someone make .NET Core better with a simple global command? Fanie Reynders did and he did it in a simple and elegant way. I'm envious, in fact, because I spec'ed this exact thing out in a meeting a few months ago but I could have just done it like he did and I would have used fewer keystrokes! Last year when .NET Core was just getting started, there was a "DNVM" helper command that you could use to simplify dealing with multiple versions of the .NET SDK on one machine. Later, rather than 'switching global SDK versions,' switching was simplified to be handled on a folder by folder basis. That meant that if you had a project in a folder with no global.json that pinned the SDK version, your project would use the latest installed version. If you liked, you could create a global.json file and pin your project's folder to a specific version. Great, but I would constantly have to google to remember the format for the global.json file, and I'd constan...

15 Web Design Trends to Watch in 2018

The modern world is full of extraordinary things that influence our imagination and mood. Our soul needs a perfect atmosphere and impressive spots. To apply such things in practice, we have submitted the list of the web trends that deserve your attention. Robert frost design analysis will meet all your wishes and expectations. Image Source Web Design Trends to Watch in 2018 1. More Organic Shapes Until this year, web design, as well as mobile design, were based on the right-angled and sharp-edged shapes. However, it seems that this year will bring some significant changes in the field of web design. The recent trends will offer the absolute rounded corners. In addition, the web design of 2018 will make the real things look like the cartoonish ones. 2.   Bold Minimalism Although some of you may think that this web design trend will not attract the Internet users. Indeed, the notion of minimalism is often associated with boredom and dullness. However, in this case, bold ...

R vs Python for Machine Learning

There are so many things to learn before to choose which language is good for Machine Learning. We will discuss each and everything about R as well as Python and the situation or problem in which situation we have to use which language. Let’s start Python and R are the two most Commonly used Programming Languages for Machine Learning and because of the popularity of both the languages Novice or you can say fresher are getting confused, whether they should choose R or Python language to commence their career in the Machine learning domain. Don’t worry guys through this article we will discuss R vs Python for Machine Learning. So, without exaggerating this article let’s get started. We will start it from the very Basics things or definitions. R vs Python for Machine Learning Introduction R is a programming language made by statisticians and data miners for statistical analysis and graphics supported by R foundation for statistical computing. R also provides high-quality graphics and...