go-redis/example/throughput

Redis Throughput Test

A simple command-line tool to test Redis client throughput with configurable parameters.

Features

• Support for both single-node and cluster Redis deployments
• Configurable read/write ratio
• Concurrent workers for parallel load generation
• Optional autopipelining support
• Real-time progress reporting
• Detailed statistics

Building

cd examples/throughput
go build

Usage

Basic Usage

Test a local Redis instance:

./throughput

Test a Redis cluster:

./throughput -mode cluster -host redis-cluster.example.com -port 6379

Command-Line Options

Flag	Default	Description
-host	localhost	Redis host
-port	6379	Redis port
-mode	single	Redis mode: single or cluster
-password	""	Redis password
-db	0	Redis database (single mode only)
-workers	10	Number of concurrent workers
-duration	10s	Test duration
-key-size	16	Size of keys in bytes
-value-size	64	Size of values in bytes
-read-ratio	0.5	Ratio of read operations (0.0-1.0)
-autopipeline	false	Enable autopipelining
-pipeline-size	100	Autopipeline batch size
-pipeline-delay	100µs	Autopipeline flush delay
-report-interval	1s	Interval for progress reports

Examples

High-throughput test with autopipelining

./throughput -workers 50 -duration 30s -autopipeline -pipeline-size 200

Write-heavy workload

./throughput -read-ratio 0.2 -workers 20 -duration 60s

Read-only workload

./throughput -read-ratio 1.0 -workers 100 -duration 30s

Large values test

./throughput -value-size 1024 -workers 10 -autopipeline

Cluster mode with authentication

./throughput -mode cluster -host redis-cluster.example.com -password mypassword -workers 50

Output

The tool provides real-time progress updates and a final summary:

=== Redis Throughput Test ===
Mode:            single
Endpoint:        localhost:6379
Workers:         10
Duration:        10s
Key size:        16 bytes
Value size:      64 bytes
Read ratio:      0.50
Autopipeline:    true
Pipeline size:   100
Pipeline delay:  100µs

✓ Connected to Redis
✓ Autopipelining enabled

Starting test...
[14:23:01] Ops: 125430 | Throughput: 125430 ops/sec | Errors: 0
[14:23:02] Ops: 251234 | Throughput: 125804 ops/sec | Errors: 0
...

=== Test Results ===
Duration:        10.002s
Total ops:       1254321
  GET ops:       627160 (50.0%)
  SET ops:       627161 (50.0%)
Errors:          0 (0.00%)
Throughput:      125407 ops/sec
Avg latency:     79.74 µs/op

Finding the Sweet Spot

Use the included script to automatically find the optimal configuration for your setup:

./find-sweet-spot.sh [host] [port]

This will test various combinations of workers, batch sizes, and delays to find the configuration that gives you maximum throughput.

Example output:

=== Best Configuration ===
Throughput: 964066 ops/sec
Config: workers=300, batch=500, delay=0

Tips for Best Performance

1. Use autopipelining for maximum throughput: -autopipeline
2. Find your sweet spot using the find-sweet-spot.sh script
3. Optimal configuration (for local Redis on M4 Max):
   • Workers: 300
   • Batch size: 500
   • Delay: 0
   • Result: ~964k ops/sec with 1.04 µs/op latency
4. Tune pipeline parameters based on your network latency:
   • Lower latency: smaller batch size, shorter delay (or 0)
   • Higher latency: larger batch size, longer delay
5. Adjust read/write ratio to match your workload: -read-ratio 0.8
6. Run on the same network as Redis for accurate results

Performance Comparison

Autopipeline vs No Autopipeline

Based on testing with local Redis on Apple M4 Max:

Workers	Without Autopipeline	With Autopipeline (optimized)	Improvement
10	100k ops/sec (10.04 µs)	110k ops/sec (9.10 µs)	1.1x
50	132k ops/sec (7.57 µs)	405k ops/sec (2.47 µs)	3.1x
100	137k ops/sec (7.32 µs)	642k ops/sec (1.56 µs)	4.7x
300	135k ops/sec (7.41 µs)	964k ops/sec (1.04 µs)	7.1x

Key Findings:

• Autopipelining provides minimal benefit with low concurrency (10 workers)
• Autopipelining provides massive benefits with high concurrency (300 workers)
• 7.1x throughput improvement at optimal configuration
• 7.1x latency reduction at optimal configuration
• Without autopipelining, throughput plateaus around 135k ops/sec regardless of worker count
• With autopipelining, throughput scales linearly with worker count up to the sweet spot

Graceful Shutdown

Press Ctrl+C to stop the test early. The tool will gracefully shut down and display results for the completed portion of the test.