Benchmarks

Detailed performance measurements. The numbers below are reproducible via scripts/run-all-benchmarks.ps1 and tracked by the CI regression harness; absolute results depend on hardware. Comparative figures against Apache Kafka are preliminary (target) until the 1.0 head-to-head publication.

Throughput

100K Messages, 100 bytes (target)

Metric Kafka Surgewave Native
Producer 68,213 msg/s 1,250,000 msg/s
Consumer 138,504 msg/s 1,282,051 msg/s

By Message Size

Size Producer Consumer
100 bytes 1.25M msg/s 1.28M msg/s
1 KB 800K msg/s 850K msg/s
10 KB 250K msg/s 280K msg/s
100 KB 50K msg/s 60K msg/s

Latency

Specific P50/P99 latency numbers are preliminary targets pending a head-to-head benchmark on identical hardware. The CI regression harness covers throughput and transport overhead; a dedicated latency-percentile project (Kuestenlogik.Surgewave.Benchmarks.Latency) will be exercised as part of the 1.0 release sign-off and the resulting figures published here.

By Storage Backend

Backend P50 Write P50 Read
Memory ultra-low (target) ultra-low (target)
ZeroCopyWal ~100 µs ~20 µs
File ~100 µs ~50 µs

Shared Memory Transport

Operation P50 P99 Throughput
Ring Buffer Write 0.1 µs 0.4 µs 625K msg/s
E2E (SHM) ultra-low (target) ~5 µs 625K msg/s

Storage Performance

Message Size File ZeroCopyWal Speedup
100 bytes 286K msg/s 549K msg/s 1.9x
1 KB 221K msg/s 235K msg/s 1.1x
10 KB 120K msg/s 145K msg/s 1.2x

Compression

Codec Compress Decompress Ratio
None - - 1.0x
LZ4 2.5 GB/s 4.0 GB/s 2-3x
Zstd 500 MB/s 1.5 GB/s 3-5x
Gzip 150 MB/s 400 MB/s 2-4x
Snappy 1.5 GB/s 2.0 GB/s 2-3x

Running Benchmarks

CLI

surgewave benchmark --messages 100000 --size 1024
surgewave benchmark -n 1000000 -s 100 --batch 1000

Embedded

// High-performance setup
await using var broker = new EmbeddedSurgewave(options =>
{
    options.Storage = StorageBackend.Memory;
    options.UseChannelPipeline = true;
});

Environment

Benchmarks run on:

  • CPU: 8 cores
  • Memory: 32 GB
  • Storage: NVMe SSD
  • .NET: 10.0

Next Steps

Edit this page