Methodology

Harness

Chabi uses the Go-based apps/membench harness derived from DiceDB's membench project.

Relevant commands already present in the repository:

• make bench
• apps/membench/Makefile
• GitHub Actions benchmark workflow in .github/workflows/bench.yml

Published Environment

The published benchmark table uses:

• Machine: Hetzner CCX23
• CPU: 4 vCPU
• Memory: 16 GB RAM

Build Mode

• Chabi benchmarked through the Redis-compatible protocol path
• Redis benchmarked through the same membench style workload
• DiceDB benchmarked in the same comparison set

Harness

Chabi uses the Go-based apps/membench harness derived from DiceDB's membench project.

Relevant repository references:

• apps/membench
• apps/membench/Makefile
• .github/workflows/bench.yml

Command Shape

The benchmark workflow in this repository uses the membench benchmark command shape:

go run main.go benchmark --database redis --host 127.0.0.1 --port <port> --num-requests <n> --num-clients <n>

Notes

• The published table compares Chabi, Redis, and DiceDB.
• The hardware reference for the published numbers is the Hetzner CCX23 machine above.
• This page documents the benchmark setup used for the published comparison rather than a local workstation run.

Performance will still vary with:

• CPU architecture
• persistence configuration
• operating system
• kernel and socket behavior
• request count and client count
• build mode