.NET 9 performance improvements

a derived Regex implementation tailored to the pattern. This avoids the interpretation overhead of the regex engine and reduces memory allocations during matching. .NET 9's generator produces tighter loops and better utilizes hardware intrinsics for character class checks.

4. GC Tuning for Server Workloads

.NET 9 improves Server GC concurrency. For latency-sensitive services, tuning GC settings can further reduce pause times.

Implementation: Configure GC settings via environment variables or runtime config for high-throughput scenarios.

{
  "configProperties": {
    "System.GC.Server": true,
    "System.GC.Concurrent": true,
    "System.GC.HeapHardLimit": 2147483648,
    "System.GC.RetainVM": false
  }
}

Rationale: GC.HeapHardLimit prevents the GC from growing beyond a defined limit, forcing more aggressive collections to stay within memory budgets. RetainVM: false returns memory to the OS, which is beneficial in containerized environments where memory limits are enforced by cgroups. .NET 9's concurrent marking runs more efficiently, reducing the suspension time required for heap compaction.

5. Hardware Intrinsics and Vectorization

.NET 9 extends support for ARM64 and AVX-512 intrinsics. The JIT automatically vectorizes loops where possible, but explicit intrinsics can be used for critical math paths.

Implementation: Ensure the runtime can detect hardware capabilities. Use System.Runtime.Intrinsics for custom vectorization.

using System.Runtime.Intrinsics;
using System.Runtime.Intrinsics.X86;

public static class MathOps
{
    public static void AddVectors(float[] a, float[] b, float[] result)
    {
        if (Avx.IsSupported)
        {
            int vectorSize = Vector256<float>.Count;
            int i = 0;
            for (; i <= a.Length - vectorSize; i += vectorSize)
            {
                Vector256<float> va = Avx.LoadVector256(&a[i]);
                Vector256<float> vb = Avx.LoadVector256(&b[i]);
                Avx.Store(&result[i], Avx.Add(va, vb));
            }
            // Handle remainder
            for (; i < a.Length; i++) result[i] = a[i] + b[i];
        }
        else
        {
            // Fallback
            for (int i = 0; i < a.Length; i++) result[i] = a[i] + b[i];
        }
    }
}

Rationale: Explicit intrinsics allow processing multiple data points per CPU cycle. .NET 9 improves the JIT's ability to auto-vectorize simple loops, but manual intrinsics remain necessary for complex algorithms. This approach scales performance linearly with SIMD width, providing massive throughput gains for data processing workloads.

Pitfall Guide

Upgrading to .NET 9 and optimizing performance introduces specific risks. Avoid these common mistakes to ensure stability and maintainability.

1. Blind Native AOT Adoption

   • Mistake: Enabling Native AOT on applications with heavy reflection or dynamic code generation.
   • Consequence: Runtime crashes or severe performance degradation due to trimmer removing required types.
   • Best Practice: Audit dependencies for reflection usage. Use PublishAot only for services with static call graphs or those using explicit AotCompatible libraries. Validate with dotnet publish warnings.

2. Ignoring GC Heap Limits in Containers

   • Mistake: Setting GCHeapHardLimit without accounting for container memory limits.
   • Consequence: OOM kills by the container orchestrator when the GC cannot reclaim memory fast enough.
   • Best Practice: Set GCHeapHardLimit to approximately 70-80% of the container's memory limit to leave headroom for native allocations and GC overhead.

3. Misusing Span<T> with Managed Arrays

   • Mistake: Holding onto a Span<T> derived from a managed array across asynchronous boundaries or thread switches.
   • Consequence: Memory corruption or access violations as the GC may move the array.
   • Best Practice: Span<T> is stack-only and cannot escape async methods. Use Memory<T> or ArrayPool<T> for cross-async scenarios. Ensure spans are used only within synchronous, short-lived scopes.

4. Over-Optimizing with Intrinsics

   • Mistake: Writing manual intrinsics for logic that the JIT can already vectorize.
   • Consequence: Increased code complexity, maintenance burden, and potential performance regression on architectures without the specific intrinsics.
   • Best Practice: Profile first. Rely on JIT auto-vectorization for standard loops. Use intrinsics only when profiling identifies a bottleneck that the JIT cannot resolve.

5. Neglecting Third-Party Library Compatibility

   • Mistake: Upgrading the SDK while using libraries that are not .NET 9 compatible or optimized.
   • Consequence: Build failures or runtime errors. Libraries may not benefit from .NET 9 improvements if they target older TFMs.
   • Best Practice: Verify all NuGet packages target net9.0 or are compatible. Update dependencies before upgrading the runtime. Check package repositories for .NET 9 specific updates.

6. Disabling Tiered Compilation Incorrectly

   • Mistake: Disabling tiered compilation to force optimization, increasing startup time unnecessarily.
   • Consequence: Slower cold starts without significant throughput gains for short-lived processes.
   • Best Practice: Keep tiered compilation enabled for most workloads. Use DOTNET_TieredPGO to enable Profile-Guided Optimization for long-running services where peak throughput is critical.

7. Skipping Benchmarking Post-Upgrade

   • Mistake: Assuming performance improvements without validation.
   • Consequence: Missing regressions in specific code paths or failing to realize expected gains.
   • Best Practice: Run benchmark suites (e.g., BenchmarkDotNet) against .NET 8 and .NET 9. Compare metrics for critical paths. Use dotnet-trace and dotnet-counters to validate GC and JIT behavior.

Production Bundle

Action Checklist

• Upgrade SDK: Install .NET 9 SDK and update global.json to pin the version.
• Update TFMs: Change project target frameworks to net9.0 and resolve build warnings.
• Audit Dependencies: Verify all NuGet packages are compatible with .NET 9 and update to latest versions.
• Enable Native AOT: Identify suitable services (APIs, workers) and enable PublishAot with trimming.
• Configure GC: Review GC settings for server workloads; apply GCHeapHardLimit and RetainVM as needed.
• Optimize Hot Paths: Refactor collection access using CollectionsMarshal and replace runtime regex with source generators.
• Benchmark: Run performance tests comparing .NET 8 vs .NET 9; validate throughput and latency metrics.
• Deploy Canary: Roll out to a subset of traffic; monitor error rates, memory usage, and P99 latency.

Decision Matrix

Configuration Template

csproj for High-Performance Native AOT Service:

<Project Sdk="Microsoft.NET.Sdk.Web">
  <PropertyGroup>
    <TargetFramework>net9.0</TargetFramework>
    <OutputType>Exe</OutputType>
    <PublishAot>true</PublishAot>
    <PublishTrimmed>true</PublishTrimmed>
    <InvariantGlobalization>true</InvariantGlobalization>
    <StripSymbols>true</StripSymbols>
    <EnableTrimAnalyzer>true</EnableTrimAnalyzer>
    <IlcOptimizationPreference>Speed</IlcOptimizationPreference>
  </PropertyGroup>
  
  <ItemGroup>
    <!-- Ensure AOT-compatible packages -->
    <PackageReference Include="Microsoft.Extensions.Hosting" Version="9.0.0" />
  </ItemGroup>
</Project>

appsettings.json for Server GC Tuning:

{
  "configProperties": {
    "System.GC.Server": true,
    "System.GC.Concurrent": true,
    "System.GC.HeapHardLimit": 1610612736,
    "System.GC.RetainVM": false,
    "System.GC.NoAffinitize": true,
    "System.GC.HeapCount": 0
  }
}

Quick Start Guide

1. Install .NET 9 SDK:

   # Linux/macOS
   curl -sSL https://dot.net/v1/dotnet-install.sh | bash /dev/stdin --channel 9.0
   # Windows
   winget install Microsoft.DotNet.SDK.9
   

2. Create Optimized Project:

   dotnet new webapi -n PerfApi --use-minimal-apis
   cd PerfApi
   dotnet add package BenchmarkDotNet
   

3. Enable Native AOT: Edit PerfApi.csproj, add <PublishAot>true</PublishAot> and <InvariantGlobalization>true</InvariantGlobalization>.

4. Publish and Test:

   dotnet publish -c Release -o ./publish
   time ./publish/PerfApi
   

   Measure startup time and memory usage. Compare against a non-AOT build to validate gains.

5. Benchmark Hot Path: Add a benchmark using BenchmarkDotNet to measure JSON serialization throughput. Run with dotnet run -c Release and analyze the report for .NET 9 improvements.