Benchmark Results

April 26, 2026 · View on GitHub

Updated: 2026-04-25
ActualLab.Fusion Version: 12.3.76

This page summarizes benchmark results from ActualLab.Fusion.Samples repository.

Test Environment

ComponentSpecification
CPUAMD Ryzen 9 9950X3D 16-Core Processor
RAM96 GB DDR5
OSWindows 11
.NET Version10.0.7

Note that Ryzen 9 9950X3D has 32 logical cores due to SMT.

Reference Redis Benchmark

Reference benchmark using redis-benchmark tool on the same machine (500K requests, best of 5 runs). Optimal client count (12) was determined via binary search over 1-1000 range.

OperationResult
PING_INLINE231.59K req/s
GET229.25K req/s
SET229.67K req/s

Run-Benchmark.cmd

The benchmark measures throughput of a simple repository-style user lookup service that retrieves and updates user records from a database: UserService.Get(userId) and Update(userId, ...).

To run the benchmark:

dotnet run -c Release --project src/Benchmark/Benchmark.csproj --no-launch-profile

Local Services

TestResultSpeedup
Regular Service118.15K calls/s
Fusion Service261.32M calls/s~2,212x

Remote Services

TestResultSpeedup
HTTP Client → Regular Service80.43K calls/s
HTTP Client → Fusion Service393.65K calls/s~4.9x
ActualLab.Rpc Client → Fusion Service7.92M calls/s~98x
Fusion Client → Fusion Service215.45M calls/s~2,679x

Run-RpcBenchmark.cmd

This benchmark compares ActualLab.Rpc with gRPC, SignalR, and other RPC frameworks. The tables below include only ActualLab.Rpc, gRPC, and SignalR. Other options, such as StreamJsonRpc and RESTful API, are way slower, so we omit them.

There are two benchmarks in RpcBenchmark project:

RPC calls:

dotnet run -c Release --project src/RpcBenchmark/RpcBenchmark.csproj --no-launch-profile -- test -b calls -l rpc,grpc,signalr -f msgpack6c -s -n 6

RPC streaming:

dotnet run -c Release --project src/RpcBenchmark/RpcBenchmark.csproj --no-launch-profile -- test -b streams -l rpc,grpc,signalr -f msgpack6c -s -n 6

Calls

TestActualLab.RpcgRPCSignalR
Sum10.16M calls/s1.29M calls/s5.31M calls/s
GetUser9.03M calls/s1.26M calls/s4.43M calls/s
SayHello6.16M calls/s1.18M calls/s2.24M calls/s

Call Latency

TestActualLab.Rpc (p50 / p95 / p99)gRPC (p50 / p95 / p99)SignalR (p50 / p95 / p99)
Sum2.0ms / 2.7ms / 6.4ms3.3ms / 5.1ms / 13.7ms3.9ms / 8.1ms / 11.1ms
GetUser2.2ms / 3.1ms / 8.7ms3.3ms / 5.9ms / 16.7ms4.6ms / 5.8ms / 11.1ms
SayHello3.2ms / 4.3ms / 10.4ms3.5ms / 4.7ms / 14.2ms9.0ms / 12.8ms / 15.0ms

Streams

Test names indicate item size: Stream1 = 1-byte items, Stream100 = 100-byte items, Stream10K = 10KB items.

TestActualLab.RpcgRPCSignalR
Stream196.96M items/s43.78M items/s18.30M items/s
Stream10043.01M items/s25.87M items/s14.25M items/s
Stream10K820.08K items/s572.76K items/s414.72K items/s

Throughput

Throughput is computed as items/s × item_size. Stream10K uses 10KB = 10,240 bytes.

TestActualLab.RpcgRPCSignalR
Stream196.96 MB/s43.78 MB/s18.30 MB/s
Stream1004.30 GB/s2.59 GB/s1.43 GB/s
Stream10K8.40 GB/s5.86 GB/s4.25 GB/s

Docker-Based RPC Benchmarks

These benchmarks run in Docker containers with CPU limits to measure 4-core server performance. The server container is limited to 4 CPUs while client containers have 24 CPUs available, ensuring the server is the bottleneck. This setup matches grpc_bench, SayHello w/ gRPC is identical to what grpc_bench measures.

Docker Calls

FrameworkSumGetUserSayHello
ActualLab.Rpc4.75M calls/s4.38M calls/s2.52M calls/s
SignalR2.23M calls/s1.82M calls/s842.45K calls/s
gRPC437.85K calls/s441.44K calls/s399.32K calls/s
MagicOnion392.59K calls/s402.85K calls/s362.84K calls/s
StreamJsonRpc265.72K calls/s226.77K calls/s99.09K calls/s
HTTP105.25K calls/s103.12K calls/s88.18K calls/s

Docker Call Latency

FrameworkSum (p50 / p95 / p99)GetUser (p50 / p95 / p99)SayHello (p50 / p95 / p99)
ActualLab.Rpc3.8ms / 8.1ms / 23.7ms4.3ms / 7.6ms / 16.2ms6.5ms / 35.5ms / 39.9ms
SignalR5.2ms / 12.2ms / 50.7ms7.0ms / 12.0ms / 37.3ms19.9ms / 51.0ms / 58.1ms
gRPC3.3ms / 32.3ms / 45.4ms3.5ms / 6.8ms / 32.1ms4.2ms / 9.6ms / 36.2ms
MagicOnion4.5ms / 8.3ms / 24.1ms5.0ms / 10.8ms / 23.1ms5.5ms / 8.8ms / 12.3ms
StreamJsonRpc43.4ms / 56.4ms / 59.6ms58.9ms / 70.1ms / 72.5ms107.3ms / 212.3ms / 222.0ms
HTTP33.9ms / 51.4ms / 54.9ms34.2ms / 44.3ms / 45.9ms30.4ms / 44.0ms / 45.5ms

Docker Streams

Test names indicate item size: Stream1 = 1-byte items, Stream100 = 100-byte items, Stream10K = 10KB items.

FrameworkStream1Stream100Stream10K
ActualLab.Rpc31.80M items/s12.66M items/s279.72K items/s
gRPC11.27M items/s6.04M items/s125.64K items/s
SignalR5.42M items/s3.62M items/s106.20K items/s
StreamJsonRpc115.20K items/s115.20K items/s0 items/s