Skip to content

README.md

Latest commit

 

History

History
126 lines (88 loc) · 3.46 KB

README.md

File metadata and controls

126 lines (88 loc) · 3.46 KB

Write Batching Benchmark

This benchmark demonstrates hint-based write batching performance across different throughput levels.

Overview

The benchmark runs write operations at four different target throughput levels with corresponding batching hints:

  • 1,000 ops/sec (baseline - no batching, batch:0)
  • 10,000 ops/sec (low latency batching, batch:1)
  • 100,000 ops/sec (moderate batching, batch:10)
  • 1,000,000 ops/sec (aggressive batching, batch:100)

For each load level, it tracks:

  • Batching window (ms)
  • Actual throughput (ops/sec)
  • Average latency (ms)

Running the Benchmark

cd benchmarks/batching
go run main.go

This will:

  1. Run benchmarks at each load level (3 seconds each)
  2. Generate bars_postgres.dat and bars_mysql.dat with raw data
  3. Generate plot_bars.gnu gnuplot script

Generating the Graph

Requires gnuplot:

# Install gnuplot (if needed)
# Ubuntu/Debian: apt install gnuplot
# macOS: brew install gnuplot
# Arch: pacman -S gnuplot

# Generate the graph
gnuplot plot_bars.gnu

This creates throughput_latency_comparison.png showing:

  • Throughput (ops/sec) achieved at each target rate
  • Latency (ms) for each batching configuration
  • Side-by-side comparison of PostgreSQL and MariaDB

Expected Behavior

Baseline (1k ops/sec, batch:0)

  • No batching - immediate execution
  • Lowest throughput baseline
  • Lowest latency
  • Demonstrates performance without batching overhead

Low Latency (10k ops/sec, batch:1)

  • 1ms batching window
  • Good balance of low latency and some batching benefits
  • Minimal additional latency from batching

Moderate Batching (100k ops/sec, batch:10)

  • 10ms batching window
  • Significant batching efficiency gains
  • Moderate latency increase
  • Good for high-throughput scenarios

Aggressive Batching (1M ops/sec, batch:100)

  • 100ms batching window
  • Maximum batching efficiency
  • Higher latency but massive throughput gains
  • Best for extreme write loads where latency is less critical

Interpreting Results

The graph shows hint-based batching performance:

  1. Throughput: How many ops/sec achieved at each target rate
  2. Latency: The trade-off in latency for batching gains
  3. Predictability: Consistent performance based on hint values
  4. Control: Explicit control over batching behavior per query

Configuration

The benchmark uses these batching configurations:

Target Rate Batch Hint Purpose
1k ops/sec batch:0 Baseline (no batching)
10k ops/sec batch:1 Low latency batching (1ms window)
100k ops/sec batch:10 Moderate batching (10ms window)
1M ops/sec batch:100 Aggressive batching (100ms window) 
Config{
    MaxBatchSize: 1000,  // Maximum operations per batch
}

Files Generated

  • bars_postgres.dat - PostgreSQL performance data
  • bars_mysql.dat - MariaDB performance data
  • plot_bars.gnu - Gnuplot script for visualization
  • throughput_latency_comparison.png - Visual graph (after running gnuplot)

Understanding the Graph

The graph shows two panels (PostgreSQL and MariaDB), each displaying:

Blue bars: Throughput (k ops/sec) - actual achieved rate
Red bars: Latency (ms) - average latency per operation

X-axis labels show the target rates (1k, 10k, 100k, 1M) with their corresponding batch hints, making it easy to see the latency/throughput trade-off at each level.