Lavik

LAVIK 0.2.0 · PUBLISHED BENCHMARKS

NVMe SSD capacity. Measured performance.

From a 2-vCPU instance to a billion-key dataset: explore throughput, tail latency, and complete reproduction procedures. Choose the experiment that matches your workload.

4 topics · 3 reports2026-10-21 / 22NVMe SSD · SPDK / io_uring

01 / Redis & Valkey: Compact Instances

Lavik vs Redis & Valkey on Compact Instances

Compare Lavik, Redis 8.10.2 and Valkey 9.1.2 at 1M keys on a 2-vCPU, 16 GiB VM, then explore Lavik’s NVMe SSD capacity at 100M keys on the same instance.

95.37 GiBValue payload on a 16 GiB VM

At 100M keys, Lavik reaches 215,644 peak GET ops/s and 231,361 peak SET ops/s, retaining 100.0% and 96.6% of its 1M-key peaks. The 95.37 GiB logical value payload is about six times nominal RAM.

Charts, results & reproduction →

02 / Redis & Valkey: High Throughput

Lavik vs Redis & Valkey for High Throughput

Compare Lavik 0.2.0 SPDK on NVMe SSD with Redis 8.10.2 and Valkey 9.1.2 on DRAM at 10M keys using a 16-vCPU VM, plus a separate billion-key Lavik capacity sweep.

1,156,95610M-key GET QPS · p99.9 7.103 ms

Lavik reaches 1,156,956 peak GET QPS and 945,384 peak SET QPS at 10M keys. The separate 1B-key sweep reaches 1,129,597 GET QPS and 910,356 SET QPS.

Charts, results & reproduction →

03 / Redis-Compatible Databases on SSD

Lavik vs Redis-Compatible Databases on SSD

Compare Lavik with Dragonfly, Garnet, Apache Kvrocks, Pika and Tendis using their SSD-backed storage paths: one billion keys, a 1.024 TB raw value payload, matched concurrency and p99.9 curves.

1,129,597Lavik peak GET QPS

Lavik records the highest GET and SET sweep peaks: 1,129,597 and 910,356 QPS. At 320 connections, it delivers 652,881 GET QPS at 0.943 ms p99.9 and 701,241 SET QPS at 0.959 ms.

Charts, results & reproduction →

04 / Lavik vs Aerospike

Lavik vs Aerospike

100M records, ten fields per record, four YCSB workload mixes, and both a 100K target rate and unlimited throughput.

+24.08%Unlimited D-Uniform total throughput

Lavik’s unlimited total throughput is 9.06%–24.08% higher across the four measured workloads. Read-only C reaches 482,358 QPS with 1.754 ms READ p99.9.

Charts, results & reproduction →

Choose the right comparison

Capacity: the 2-vCPU report compares Lavik at 1M and 100M keys; Redis and Valkey controls use 1M keys.

Latency: the new reports use client-observed whole-window p99.9. Read it alongside connection count or target rate and the corresponding throughput.

Method: each point or window is measured once, without confidence intervals. Cache, persistence, CPU and warmup conditions are documented per report.

Download the tested release →