Skip to content

scripts

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

scripts

README.md

MVP Actor System - Benchmarks

This directory contains benchmark scripts to test the actor system scalability.

Available Scripts

1. benchmark.exs - Single-Node Benchmark

Tests performance on a single node.

mix run scripts/benchmark.exs

Included tests:

  • Creation of 50,000 actors
  • 10,000 invocations on the same actor (hot spot)
  • Hash ring distribution
  • Multi-capability
  • Mixed load
  • Latency measurement (P50, P95, P99)

2. benchmark_multinode.exs - Multi-Node Benchmark

Tests distributed performance with multiple Erlang nodes.

NOTE This benchmark needs to be executed in distributed mode:

elixir --name bench@127.0.0.1 --cookie mvp -S mix run scripts/benchmark_multinode.exs

DO NOT execute with just mix run, use the complete command above!

Included tests:

  • Distributed actor creation
  • RPC latency between nodes
  • Distributed mixed load
  • Resilience (node failure simulation)
  • Hash ring balancing

3. test_multinode.exs - Simple Multi-Node Test

Basic test to verify connectivity between nodes.

elixir --name test@127.0.0.1 --cookie mvp -S mix run scripts/test_multinode.exs

4. benchee_benchmark.exs - Professional Benchmarks with Benchee

Statistical benchmarks using the Benchee library with extended metrics.

mix run scripts/benchee_benchmark.exs

Included tests:

  • Actor creation performance (100, 1k, 5k actors)
  • Invocation performance (sequential vs parallel)
  • Multi-capability comparison
  • Hot spot contention analysis
  • Hash ring distribution patterns

Features:

  • Extended statistics (median, percentiles, standard deviation)
  • Memory profiling per operation
  • Comparison mode between scenarios
  • Warmup and multiple iterations for accuracy

Expected Warnings

:slave module warnings

warning: :slave.start/3 is deprecated

This is normal. The :slave module will be replaced by :peer in OTP 29. For now, it still works perfectly.

Errors during cluster initialization

[error] Process #PID<...> raised an exception
** (ErlangError) Erlang error: {:exception, {:undef, [{Capabilities, :_update_capabilities_from_remote...

This is normal and expected. During cluster initialization, nodes connect before the application is fully loaded, causing RPC attempts that fail. These errors are captured and do not affect system functionality.

Why it happens:

  1. Node A and Node B connect
  2. Capabilities on Node A detects nodeup from Node B
  3. Tries to propagate capabilities via RPC
  4. But the application is not yet loaded on Node B
  5. RPC fails with :undef (function not found)
  6. System ignores and continues normally

After complete initialization, the system works perfectly without errors.

Expected Results

Single-Node (85,000 actors)

  • Creation throughput: ~16,000 actors/s
  • Invocation throughput: ~30,000 req/s
  • P95 latency: < 100μs
  • Memory per actor: ~11 KB
  • Success rate: 100%

Multi-Node (4 nodes, ~10,000 actors)

  • Creation throughput: ~8,000 actors/s
  • Distributed throughput: ~8,000 req/s
  • RPC latency: ~150μs
  • Balanced distribution across nodes
  • Success rate: >99%