Skip to content

README.md

Latest commit

 

History

History
144 lines (89 loc) · 4.07 KB

README.md

File metadata and controls

144 lines (89 loc) · 4.07 KB

Validators DAO Performance Testing Tools

This repository contains practical performance testing tools maintained by Validators DAO.

The purpose of this repository is simple:
to allow anyone to measure, verify, and understand the performance characteristics of Linux nodes in a reproducible way.

The tools here are designed for:

  • VPS
  • Bare metal servers
  • Cloud instances

They are intended for verification and comparison, not for marketing claims.

All tools live under tools/<tool_name>/.

Quick Start

If you just want to run a performance check on a Linux node, start with node_bench.

node_bench

node_bench is a reproducible benchmark that measures CPU, memory (optional), and disk performance at the node level.

Run directly via curl (defaults: --fio-dir /var/tmp, --fio-size-gb 32, --runtime-sec 60, --ramp-sec 10):

curl -fsSL https://storage-for-testing.erpc.global/tools/node_bench.sh | bash

If the root disk is tight (≈10–20GB free), shrink the fio file:

curl -fsSL https://storage-for-testing.erpc.global/tools/node_bench.sh | bash -s -- --fio-size-gb 4

If you want a longer disk run on a large NVMe volume, bump the size:

curl -fsSL https://storage-for-testing.erpc.global/tools/node_bench.sh | bash -s -- --fio-size-gb 64

--fio-dir already defaults to /var/tmp; only set it when you need a different mount point.

The script:

  • prints every command it executes
  • logs all output to disk
  • stops immediately if requirements are not met (permissions, disk space, paths)

Tools

geyserbench-rs

A source mirror of the geyserbench-rs benchmark tool for Solana gRPC-compatible data feeds.

Contents:

  • Rust source tree mirrored under tools/geyserbench-rs/
  • configuration, protobuf definitions, and benchmark logic as maintained upstream

This repository currently publishes node_bench directly to R2. geyserbench-rs is included here first as source so it can be reviewed, referenced, and wired into a dedicated publish flow later.

node_bench

A node-level benchmark for Linux systems.

What it measures:

  • CPU Using sysbench with a multi-thread sweep.

  • Memory Using STREAM (auto-installed or built from source; skip with --allow-missing-stream if you accept missing memory results).

  • Disk Using fio with:

    • direct I/O (direct=1)
    • fixed, explicit workloads
    • JSON output for independent analysis

Results are written to a timestamped directory:

$HOME/results/<hostname>_<UTC timestamp>/   # override with RESULTS_DIR if needed

Key files:

  • summary.txt — full console log (canonical record)
  • fio_*.json — raw fio results

Documentation:

  • See tools/node_bench/README.md for detailed usage and result interpretation.

Execution (maintained script):

Community:

  • Join the Validators DAO official Discord for active benchmarking chatter, run logs, and Q&A: https://discord.com/invite/C7ZQSrCkYR
  • Pull requests with new run results or examples are welcome.

Viewing Results

After node_bench finishes, you can inspect results immediately.

Show the latest summary:

RESULTS_BASE=${RESULTS_DIR:-$HOME/results}; cat "$(ls -td "${RESULTS_BASE}"/* | head -n1)"/summary.txt

Page through it:

RESULTS_BASE=${RESULTS_DIR:-$HOME/results}; less "$(ls -td "${RESULTS_BASE}"/* | head -n1)"/summary.txt

Show only extracted fio metrics:

RESULTS_BASE=${RESULTS_DIR:-$HOME/results}; grep -n "extracted metrics" -A20 "$(ls -td "${RESULTS_BASE}"/* | head -n1)"/summary.txt

The script also prints the result directory path to the console when it completes.

Design Principles

All tools in this repository follow the same principles:

  • no silent execution
  • no hidden fallbacks
  • no cache-based shortcuts
  • fail fast with clear errors
  • safe defaults (no destructive operations)

If a condition is not suitable for accurate measurement, the tool will stop instead of producing misleading results.

License

Apache License 2.0