Object allocation profiling

Creating objects (allocating memory in the heap) is an expensive operation. Therefore, creating a large number of temporary objects can significantly slow down a .NET application and add an extra load on the garbage collector. The .NET runtime will have to perform garbage collection more frequently, causing pauses in execution, which will ultimately negatively impact application performance.

Goals of object allocation profiling
How to control object allocation profiling
Allocations tab
Allocation profiling settings and overhead
Best practices

Goals of object allocation profiling

Identify parts of the code that generate the most temporary objects and put a strain on the garbage collector.
Allocation profiling helps with fixing memory leaks. Additional information about where a specific object was created helps answer why it still exists in memory.

How to control object allocation profiling

You can start and stop object allocation profiling at any time while your application is running. When object allocations are not recorded, the profiler adds no overhead to the application.

Profiler UI

When the profiler is connected to an application, the toolbar shows these allocation profiling controls:

Toolbar button	Description
	Start/Stop allocation profiling.
	Clear allocation profiling results and continue profiling.

Agent startup options

It may be useful to launch the application with object allocation profiling right from the start using profiler agent startup options.

HTTP API

If you need remote control or automation in your profiling workflow, YourKit .NET Profiler lets you manage allocation profiling using HTTP API endpoints:

.NET API

The YourKit .NET Profiler also provides a .NET API for deep integration with .NET applications. The YourKit.Profiler.Api.Controller includes methods for starting, stopping, and resetting the allocation profiling, similar to the toolbar buttons.

Allocations tab

The results of object allocation profiling can be analyzed from different perspectives:

Telemetry charts.
Class - Allocated objects grouped by their class.
Allocation tree - All threads - Displays methods that allocated objects, organized into a single combined call tree across all threads.
Allocation tree - By thread - Displays methods that allocated objects, organized into separate call trees per .NET thread.
Flame graph - Displays methods that allocated objects as a flame graph.
Hot spots by object count - Methods allocating the highest number of objects.
Hot spots by object size - Methods allocating the most memory.
Method list - Provides a complete list of all methods responsible for object allocations.

Allocation profiling settings and overhead

Profiler overhead directly depends on the number of objects for which the profiler records a stack trace. The fewer stack traces the profiler records, the lower the overhead.

YourKit .NET Profiler allows you to control overhead using the sampling interval parameter. For each .NET thread, the profiler maintains a counter that tracks the number of bytes allocated by that thread. When the counter exceeds the sampling interval during object creation, the profiler records an allocation sample.

The larger the sampling interval, the fewer samples are recorded and the lower the overhead. Conversely, the smaller the sampling interval, the higher the overhead. If the interval is set to 1, the profiler records allocation data for every object.

Best practices

Choose the right profiling options

Choose reasonable profiling settings to keep profiling overhead moderate. In most cases, it is not necessary to track the creation location of every object.

Limit profiling duration

Since object allocation profiling adds overhead to your .NET application, we recommend not keeping this mode enabled all the time. For example, if you want to profile object allocations when your web application processes an HTTP request, the optimal approach would be:

Start allocation profiling just before executing the request.
Execute the request, take a snapshot, and stop allocation profiling.
Analyze the data in the snapshot.

With this approach, you can minimize performance impact and even profile in a production environment.

Validate changes

Always baseline performance before changes and re-measure after.