With the annual November release of new .NET versions, Stephen Toub, a Partner Software Engineer on the .NET team, publishes an in-depth analysis of performance improvements around September each year. Given the yearly release cycle, I have decided to update my performance book exclusively for the Long-Term Support (LTS) versions. To complement the book, I will publish articles covering the performance changes in Short-Term Support (STS) versions like .NET 9.

In this article, I will highlight the major performance differences between these two versions.

Performance Improvements & Slowdowns

Below is a brief overview of some key performance enhancements and potential slowdowns introduced in .NET 9. I found some of these results particularly surprising, especially in areas like character handling, integers, doubles, object creation, string comparison, and string compression.

CATEGORY	DESCRIPTION	PERFORMANCE RESULT
Array	All() with a predicate for reference types	Increase of 1.6 times
Array	Count() with a predicate for reference types	Increase of 2.28 times
Array	foreach() for reference types	Decrease of 1.11 times
Array	LongCount() for reference types	Increase of 1.17 times
Array	Parallel.For() for value types	Decrease of 1.6 times
Array	Parallel.For(): + MaxDegreeOfParallelism for value types	Decrease of 1.56 times
Array	Parallel.ForEach() + MaxDegreeOfParallelism  for value types	Decrease of 1.6 times
Array	Parallel.ForEach() for value types	Decrease of 1.6 times
Character	EndsWith()	Decrease of 23.4 times
Character	StartsWith()	Increase of 36.75 times
Cloning	Reference type	Increase of 1.14 times
Cloning	Value type	Increase of 1.12 times
Code	Environment.ProcessId	Increase of 1.43 times
Code	Null coalescing	Increase of 1.16 times
Code	Null coalescing assignment using ??	Increase of 1.2 times
Dictionary	ContainsKey()	Increase of 1.1
Dictionary	Convert to ImmutableDictionary	Increase of 1.4 times
Dictionary	Iterating using for()	Increase of 1.24
Exceptions	Trapping exceptions using the when clause	Increase of 2.11 times
Exceptions	Trapping exceptions using try/catch	Increase of 2.2 times
ImmutableDictionary	Iterating using for()	Increase of 1.68
ImmutableSortedDictionary	Iterating using for()	Increase of 2.44
IQueryable	Any() for reference type	Increase of 2 times
JsonSerializer	Deserialize with JsonSerializerContext	Increase of 1.28 times
List	All() with a predicate for reference type	Increase of 2 times and allocation decreased by half
List	All() with a predicate for value type	Increase of 4.19 times and allocations decreased by almost 100 bytes
List	Any() for reference type	Increase of 1.44 times
List	Any() for value type	Increase of 2.48 times
List	Any() with a predicate for reference type	Increase of 3.14 times and allocations decreased by almost half
List	Any() with a predicate for value type	Increase of 1.82 times
List	Convert to ImmutableDictionary	Increase of 1.4 times
List	Count() for value type	Increase of 2.36 times
List	Count() with a predicate for reference type	Increase of 2.43 times and allocations decreased by almost half
List	Count() with a predicate for value type	Increase of 4.39 times
List	foreach with CollectionsMarshall.AsSpan() for value types	Increase of 1.23 times
List	ForEach() for value types	Increase of 1.31 times
List	foreach() for value types	Increase of 1.35 times
List	Iterating using for() for value types	Increase of 1.28 times
List	LINQ First()	Increase of 7.6 times
List	LongCount() for value type	Increase of 2.44 times
Method	in params	Increase of 1.19 times and decrease of allocations by 15,559 bytes
Method	params normal	Increase of 1.21 times and decrease of allocations by 16,119 bytes
Method	ref readonly params	Increase of 1.19 times and decrease of allocations by 17,271 bytes
Objects	Assigning to a variable for an integer return	Increase of 8.33 times
Objects	Create value type	Increase of 1.37 times
Objects	CreateInstance() for a value type	Increase of almost 2 times
Objects	Creating a public class	Increase of 4 times
Objects	Creating a public sealed class	Increase of 5 times
Objects	Creating a value object	Increase of 1.63 times
Objects	Creating an internal class	Increase of 4 times
Objects	Creating an internal sealed class	Increase of 4.44 times
Objects	Creating Attribute	Increase of 3.7 times
Objects	Creating sealed Attribute	Increase of 3.86 times
Objects	Disposing of objects using the new style of using	Increase of 1.2 times
Objects	Disposing of objects using the using statement	Increase of 1.18 times
Objects	Disposing of objects using try/finally	Increase of 1.17 times
Objects	GetUninitializedObject() for reference type	Increase of 4.32 times
Objects	GetUnitializedObject for reference type	Increase of 4.32 times
Objects	Using discard for integer return	Decrease of 15.3 times
Span	Fill()	Decrease of 2.12 times
String	Combine 2 strings using Join()	Increase of 1.45 times
String	Comparison using AsSpan().Equals(OrdinalIgnoreCase)	Increase of 22 times
String	Comparison using Equals()	Decrease of 6.32 times
String	Comparison with EndsWith()	Decrease of 1.2 times
String	Contains()	Increase of 5.27 times
String	EndsWith()	Decrease of 1.29 times
String	Replace() with OrdinalIgnoreCase	Increase of 1.22 times
String Compression	Using DeflateStream async with CompressionLevel.Fastest	Decrease of 2.27 times
String Compression	Using DeflateStream async with CompressionLevel.NoCompression	Decrease of 2 times
String Compression	Using DeflateStream async with CompressionLevel.Optimal	Decrease of 2.5 times
String Compression	Using DeflateStream async with CompressionLevel.SmallestSize	Decrease of 2.55 times
String Compression	Using GZipStream async with CompressionLevel.Fastest	Decrease of 2.25 times
String Compression	Using GZipStream async with CompressionLevel.NoCompression	Decrease of 4 times
String Compression	Using GZipStream async with CompressionLevel.Optimal	Decrease of 2.5 times
String Compression	Using GZipStream async with CompressionLevel.SmallestSize	Decrease of 2.54 times
String Compression	Using ZLibStream async with CompressionLevel.Fastest	Decrease of 2.3times
String Compression	Using ZLibStream async with CompressionLevel.NoCompression	Decrease of 4.1 times
String Compression	Using ZLibStream async with CompressionLevel.Optimal	Decrease of 2.5 times
String Compression	Using ZLibStream async with CompressionLevel.SmallestSize	Decrease of 2.6 times
String Decoding	Using BigEndianUnicode	Decrease of 1.5 times
String Encoding	Using Encoding.UTF32	Decrease of 1.16 times
String Validation	Any()	Increase of 1.94 times
String Validation	Using ? and string.Length	Increase of 57 times
StringBuilder	Append()	Increase of 1.35 times
StringBuilder	AppendFormat()	Increase of 1.19 times
Type	Initializing an integer	Decrease of 7.25 times
Types	Consuming a constant that is a double	Decrease of 733.75 times
Types	Consuming a readonly property that is a double	Increase of 4.69 times
Types	Double to integer	Decrease of 3.13 times
Types	Double to integer with casting	Decrease of 3.5 times
Types	Initialize reference type	Decrease of 1.31 times and increase of allocation by 3,644 bytes
Types	Using readonly property that is a double	Decrease of 7.36 times

Summary

This article highlights only a portion of the performance changes I discovered using the thousands of benchmark tests I routinely run to track .NET improvements. I have focused on performance changes of 1.1 or higher. Overall, based on all the tests I conducted; .NET 9 is approximately 1.11 times faster than .NET 8—a notable improvement for an interim release!

This guide is intended to help you decide whether to upgrade to .NET 9. I strongly recommend upgrading as soon as possible to leverage the enhanced performance. Your users will benefit from the increased speed without any need for changes to your existing code.

Feel free to share any questions or feedback in the comments below.

Benchmark Your Code Like dotNetDave!

Pick up any books by David McCarter by going to Amazon.com: http://bit.ly/RockYourCodeBooks

If you liked this article, please buy David a cup of Coffee by going here: https://www.buymeacoffee.com/dotnetdave

© The information in this article is copywritten and cannot be preproduced in any way without express permission from David McCarter.