VPS Benchmark Script

A bash script for systematic performance testing of VPS environments. Collects CPU, memory, disk, and network metrics with historical tracking and comparison capabilities.

Features

• CPU benchmarking (single and multi-threaded via sysbench)
• Memory bandwidth testing
• Disk I/O performance using FIO (flexible I/O tester)
  • Sequential read/write tests with direct and buffered I/O
  • Automatic ioengine detection (libaio/sync)
  • Latency measurements via ioping
• Network speed tests using Ookla Speedtest CLI or Python speedtest-cli
• SQLite database for historical data storage
• Comparison mode for tracking performance changes over time
• JSON export for integration with monitoring systems
• Optional ntfy push notifications
• Dependency caching to minimize installation overhead
• System health checks (disk usage, filesystem type, load average)

Requirements

Tested on: Debian 12 (Bookworm), Ubuntu 20.04+, Fedora, RHEL/CentOS

Dependencies (auto-installed on first run):

• sysbench
• fio
• bc
• sqlite3
• curl
• ioping
• speedtest-cli (Ookla) or speedtest-cli (Python)

Root access required for initial dependency installation only.

Installation

git clone https://github.com/buildplan/vps-bm.git
cd vps-bm
chmod +x vps-bm.sh

Usage

Basic benchmark

./vps-bm.sh

Save results to database

sudo ./vps-bm.sh --save

Compare with previous run

sudo ./vps-bm.sh --compare

Quick mode (reduced test times)

sudo ./vps-bm.sh --quick --compare

Export JSON

sudo ./vps-bm.sh --json

List historical benchmarks

./vps-bm.sh --list

Command Line Options

Option	Description
-s, --save	Save results to SQLite database
-c, --compare	Save and compare with previous benchmark
-l, --list	List all saved benchmark runs
-q, --quick	Fast benchmark (5s CPU, 256MB disk)
-j, --json	Export results as JSON
-h, --help	Show help message

Configuration

Create .benchmark_config in the script directory to override defaults:

# Test durations and sizes
CPU_TEST_TIME=10          # seconds (default: 10)
DISK_TEST_SIZE="1G"       # FIO size format (default: 1G)

# Network testing
SKIP_NETWORK=0            # set to 1 to skip network tests
SPEEDTEST_SERVER_ID=""    # specific server ID, or empty for auto-select

# ntfy notifications
NTFY_ENABLED=0
NTFY_URL="https://ntfy.sh"
NTFY_TOKEN=""
NTFY_TOPIC="vps-bm"

Speedtest Server Selection

To use a specific speedtest server for consistent testing:

1. Find server ID: speedtest --servers
2. Set in config: SPEEDTEST_SERVER_ID="12345"

This ensures network tests always use the same endpoint for accurate historical comparisons.

Output Files

• benchmark_results.db - SQLite database with historical data
• benchmark_latest.json - Most recent results in JSON format
• benchmark.log - Execution log with timestamps
• .deps_installed - Dependency verification marker

Automated Monitoring

Add to crontab for weekly benchmarks:

# Weekly benchmark every Sunday at 2 AM
0 2 * * 0 /path/to/vps-bm.sh --compare >> /var/log/vps-benchmark-cron.log 2>&1

JSON Output Format

The benchmark_latest.json file follows this structure for integration with monitoring systems:

{
  "version": "0.2.0",
  "timestamp": "2025-12-07T16:40:35Z",
  "hostname": "server-name",
  "is_docker": false,
  "metrics": {
    "cpu": {
      "single_thread_events_per_sec": 1437.44,
      "multi_thread_events_per_sec": 5478.04
    },
    "memory": {
      "bandwidth_mib_per_sec": 18709.80
    },
    "disk": {
      "write_buffered_mbs": 839,
      "write_direct_mbs": 573,
      "read_mbs": 1331,
      "latency_us": 771
    },
    "network": {
      "download_mbps": 197.20,
      "upload_mbps": 188.33,
      "latency_ms": 4.37
    }
  }
}

Performance Metrics Explained

CPU

Events per second from sysbench prime number calculation (higher = better)

Memory

Sequential memory bandwidth in MiB/s from sysbench memory test

Disk (FIO)

• Write (Buffered): Sequential write with OS cache (simulates typical file operations)
• Write (Direct): Direct I/O write bypassing cache (measures raw disk performance)
• Read (Direct): Direct I/O read bypassing cache
• Latency: Average I/O latency in microseconds from ioping (lower = better)

FIO automatically selects the best I/O engine (libaio on most systems, fallback to sync).

Network

• Download/Upload: Mbps measured via Ookla Speedtest or Python speedtest-cli
• Latency: Ping latency in milliseconds

System Health Checks

The script performs pre-flight checks before benchmarking:

• Disk Usage: Warns if >90% full
• Filesystem Type: Detects tmpfs (RAM disk) and warns that disk benchmarks will be invalid
• Load Average: Reports current system load per CPU
• Docker Detection: Identifies containerized environments

If running from /tmp or a RAM disk, move the script to a physical disk location like /root or /home for accurate disk benchmarks.

Example Output

=== FINAL RESULTS SUMMARY ===

CPU Performance (sysbench):
  Single-Thread        [✓]:  1500.57 events/sec
  Multi-Thread         [✓]:  5669.13 events/sec

Memory Performance:
  Bandwidth            [✓]: 20026.91 MiB/s

Disk Performance (FIO):
  Write (Buffered)     [✓]: 807 MB/s
  Write (Direct)       [✓]: 570 MB/s
  Read (Direct)        [✓]: 1536 MB/s
  Latency              [✓]: 771 μs

Network Performance (speedtest):
  Download             [✓]: 197.20 Mbps
  Upload               [✓]: 187.74 Mbps
  Latency              [✓]: 4.37 ms

=== COMPARISON WITH PREVIOUS RUN ===
Previous Run: 2025-12-07 16:40:35 (v0.2.0)

CPU Performance:
  Single-Thread (ev/s)     : 1437.44 →  1500.57 (▲4.0%)
  Multi-Thread (ev/s)      : 5478.04 →  5669.13 (▲3.0%)

Disk Performance (MB/s):
  Write Buffered           : 839 → 807 (▼3.0%)
  Read Direct              : 1331 → 1536 (▲15.0%)

Known Limitations

• Network upload tests may occasionally timeout on congested connections - download and latency metrics are still captured
• Disk benchmarks require sufficient free space (default 1GB for tests)
• Some VPS providers limit or throttle I/O operations which may affect results
• Container environments (Docker, LXC) may show different results than bare metal due to virtualization overhead

Troubleshooting

"Missing dependencies" error Run with sudo on first execution to install required packages.

"Running in TMPFS" warning Script is in a RAM disk (/tmp or similar). Move to persistent storage: mv vps-bm.sh /root/

FIO "libaio not supported" message Normal on some systems. Script automatically falls back to sync engine.

Speedtest upload fails Common on rate-limited connections. Download and latency metrics are still recorded.

"No space left on device" Reduce DISK_TEST_SIZE in config (e.g., DISK_TEST_SIZE="512M").

Why FIO Instead of dd?

FIO (Flexible I/O Tester) provides more accurate and consistent results than dd:

• Better control over I/O patterns and direct I/O
• Less affected by OS caching
• Industry-standard tool for storage benchmarking
• More realistic simulation of real-world workloads

Contributing

Bug reports and pull requests are welcome. For major changes, open an issue first to discuss proposed modifications.

License

MIT License - see LICENSE file for details