Benchmarks

Proof

Performance that fits the edge

Leafcuttr is tuned for constrained environments, from tmpfs throughput to fsync-heavy workloads where durability still matters.

Throughput
87.64

MB/s on untuned producer workloads in tmpfs.

Durability
32.53

MB/s on tuned producer workloads with fsync enabled.

Tail latency
933 ms

p999 for Leafcuttr on native sync mode in the supplementary comparison.

Throughput comparison

Higher bars indicate stronger write throughput. Leafcuttr is shown alongside Apache Kafka across tmpfs and fsync scenarios.

Untuned Producer - Leafcuttr (tmpfs)
87.64 MB/s
Untuned Producer - AK (tmpfs)
72.1 MB/s
Tuned Producer - Leafcuttr (fsync)
32.53 MB/s
Tuned Producer - AK (fsync)
23.21 MB/s
Untuned Producer - Leafcuttr (fsync)
3.55 MB/s
Untuned Producer - AK (fsync)
1.96 MB/s

Supplementary comparison

Latency and throughput across container, native, and sync-mode runs.

Engine TP Avg Lat Max Lat p50 p999
RedPanda Docker (acks=all) 25.11 1239 3771 1169 3721
AK 4.0.0 Docker no flush 58.22 532 960 518 949
AK 4.0.0 Docker flush interval 1 8.9 3433 7733 3212 7711
AK 4.0.0 Native no flush 63.53 487 929 476 916
AK 4.0.0 Native flush interval 1 8.44 3640 6752 3644 6741
Leafcuttr Mod Native no flush 65.59 468 944 453 933
Leafcuttr Mod Native flush interval 1 9.48 3228 5158 3139 5148
Leafcuttr Mod Native Sync Mode ✦ 12.67 2413 3769 2392 3752

What this means

Leafcuttr is designed to outperform traditional Kafka in edge scenarios with better throughput, lower tail latency, and durability that still fits constrained hardware.

See Architecture → Contact →