Random I/O Tester

A comprehensive disk performance testing tool written in Rust that measures random read performance using both standard I/O and memory-mapped files.

Features

• Configurable test parameters: File count, file size, block size, operation count, thread count
• Two testing modes: Standard I/O (seek+read) vs Memory-mapped files
• Multi-threaded testing: Configurable number of concurrent threads
• Reproducible experiments: Pseudo-random with configurable seed
• First vs Repeated read tracking: Distinguishes cache effects
• Comprehensive statistics: Average, median, 90th/95th/99th percentiles, min/max
• Automatic cleanup: Test files are removed after completion

Usage

Basic Example

# Test with defaults (10 files, 1MB each, 1000 operations, 4 threads)
cargo run

# Small quick test
cargo run -- -f 3 -s 32768 -n 100 -t 2

# Large scale test
cargo run -- -f 50 -s 10485760 -n 10000 -t 8

Testing Standard I/O vs Memory-Mapped Performance

# Standard I/O mode
cargo run -- -f 10 -s 1048576 -n 1000 -t 4 --seed 42

# Memory-mapped mode (same parameters for comparison)
cargo run -- -f 10 -s 1048576 -n 1000 -t 4 --seed 42 -m

Advanced Configuration

cargo run -- \
  --num-files 20 \
  --file-size 5242880 \
  --block-size 8192 \
  --num-operations 5000 \
  --num-threads 8 \
  --wait-time 5 \
  --seed 123 \
  --use-mmap \
  --file-prefix "perftest"

Command Line Options

Option	Short	Description	Default
--num-files	-f	Number of test files to create	10
--file-size	-s	Size of each file in bytes	1048576 (1MB)
--wait-time	-w	Wait time after file creation (seconds)	1
--num-threads	-t	Number of concurrent threads	4
--seed		Random seed for reproducibility	42
--block-size	-b	Size of blocks to read (bytes)	4096
--num-operations	-n	Total number of read operations	1000
--use-mmap	-m	Use memory-mapped files	false
--file-prefix		Prefix for test file names	"testfile"

Output Interpretation

The tool reports three sets of statistics:

1. All Reads: Combined statistics for all operations
2. First Reads: Blocks read for the first time (likely from disk)
3. Repeated Reads: Blocks read again (likely from cache)

Metrics Explained

• Average/Median: Central tendency of latencies
• 90th/95th/99th percentile: Tail latency performance
• Min/Max: Best and worst case performance
• Count: Number of operations in each category

Example Output

📊 Performance Results:

📈 All Reads (1000 operations):
  Count:     1000
  Average:   15.23μs
  Median:    12.45μs
  90th %ile: 28.67μs
  95th %ile: 45.23μs
  99th %ile: 89.12μs
  Min:       3.21μs
  Max:       156.78μs

🆕 First Reads (842 operations):
  Count:     842
  Average:   16.78μs
  [... additional stats ...]

🔄 Repeated Reads (158 operations):  
  Count:     158
  Average:   8.43μs
  [... additional stats ...]

Use Cases

• Storage performance benchmarking
• Cache effectiveness analysis
• Multi-threaded I/O scalability testing
• Standard I/O vs memory-mapped performance comparison
• Reproducible performance experiments
• Tail latency analysis

Building and Running

# Build the project
cargo build --release

# Run with optimizations
cargo run --release -- [OPTIONS]

# Get help
cargo run -- --help

If you are on Linux and want to build a static executable, you can try this:

rustup target add x86_64-unknown-linux-musl
sudo apt-get install musl-tools
cargo build --target x86_64-unknown-linux-musl --release