Aaron.Akka.Benchmarker 0.1.0

actor-benchmarker

Tooling for generating automated Akka.NET end-to-end benchmarks for actors. Think big, "macro" benchmarks - not your traditional Benchmark.NET micro-benchmarks you often see on Twitter et al.

![NOTE] This library is an experiment and not meant for production use.

Benefits of Akka.Benchmarker

Why use this library?

1. Eliminates Boilerplate - we make it very easy to parameterize your macro benchmarks with a variety of different ActorSystem configuration options. We make sure to eliminate setup overhead from your benchmarks AND we cleanly tear down dependencies afterwards.
2. Forces Good Practices - these benchmark methodologies are designed by the team that build and maintains Akka.NET itself; we've had practice doing this type of benchmarking for nearly 10 years. Only now, we've codified everything into a streamlined set of abstractions that are very accessible and user-friendly.
3. No HOCON - everything in Akka.Benchmarker is configured with Akka.Hosting. No HOCON is required.
4. Parameterizable - it's very easy to test your actors under a variety of different configurations (i.e. different ActorSystem settings, serializers, Akka.Persistence back-ends).

Designing a Benchmark

The idea behind Akka.Benchmarker (the namespace for this library) is to choose the following:

1. Code Under Test: choose a single "head actor" that represents a set of related entities - in Akka.Cluster parlance this would typically be a ShardRegion. In local Akka.NET this might be the aggregate root / parent actor for something like the child-per-entity pattern. These actors typically perform complex work; have state; and are where the real "business" of your application is conducted. These actors might also interact with other dependent actors - that is totally fine: we want all of that overhead and data included in our benchmark.
2. Infrastructure Under Test: need to use SQL Server, Redis, or MongoDb as part of your test configuration? Akka.Benchmarker will allow you to fire up and tear down external infrastructure dependencies like Test Conatiners. Again, these are meant to be end to end benchmarks - real overhead is what we are trying to measure.
3. Clusters or Single Actor Systems: the IBenchmarkConfiguration construct in Akka.Benchmarker makes it very easy to say "give me a cluster of 4 nodes" or "run as a single ActorSystem" for your benchmark. Implement your own IBenchmarkConfiguration types to help express this. You can run multiple IBenchmarkConfigurations for the same code and infrastructure under test.

Implementing Your Design

TODO: going to test this with some live applications first

Copyright 2015-2024 Aaron Stannard <aaronstannard.com>