Redis - Benchmarks

Redis - Benchmarking Performance

Redis includes a built-in benchmarking utility, redis-benchmark, designed to simulate running multiple commands from multiple parallel clients. It provides a standardized way to measure the performance of your Redis instance under a synthetic load.

Syntax

The basic syntax for running the benchmark utility from your terminal is:

redis-benchmark [options]

Running a Default Benchmark

Executing redis-benchmark without any options runs a standard suite of tests using 50 parallel clients and 100,000 total requests.

$ redis-benchmark

====== PING_INLINE ======
  100000 requests completed in 0.70 seconds
  50 parallel clients
  3 bytes payload
  keep alive: 1

99.68% <= 1 milliseconds
100.00% <= 1 milliseconds
142653.36 requests per second

====== PING_BULK ======
  100000 requests completed in 0.69 seconds
  50 parallel clients
  3 bytes payload
  keep alive: 1

99.78% <= 1 milliseconds
100.00% <= 1 milliseconds
144717.80 requests per second

... (output continues for SET, GET, LPUSH, etc.) ...

====== MSET (10 keys) ======
  100000 requests completed in 1.15 seconds
  50 parallel clients
  3 bytes payload
  keep alive: 1

96.48% <= 1 milliseconds
99.88% <= 2 milliseconds
100.00% <= 2 milliseconds
87108.02 requests per second

Customizing the Benchmark

You can tailor the benchmark to better reflect your application’s workload using various options. For example, let’s test only SET and GET commands with a larger data payload.

# -n 100000: 100,000 total requests
# -c 100: 100 parallel clients
# -d 256: 256 bytes data payload for SET/GET
# -t get,set: Only run GET and SET tests
# -q: Quiet mode, shows only req/sec

$ redis-benchmark -n 100000 -c 100 -d 256 -t get,set -q

SET: 121951.22 requests per second
GET: 123001.23 requests per second

Key Benchmark Options

Below is a table of the most commonly used options for redis-benchmark.

Option	Description	Default Value
-h	Server hostname.	127.0.0.1
-p	Server port.	6379
-a	Password for authentication.	None
-c	Number of parallel connections (clients).	50
-n	Total number of requests to perform.	100000
-d	Data size of SET/GET values in bytes.	3
-t	Comma-separated list of tests to run (e.g., set,get,lpush).	All tests
-P	Pipeline requests. Use 1 to disable pipelining.	1
-k	Use persistent connections (1=yes, 0=no).	1
-r	Use random keys for SET/GET/INCR from a keyspace of size .	Not set
-q	Quiet mode. Shows only the final queries/sec summary.	Disabled
—csv	Output results in CSV format.	Disabled

Best Practices and Interpretation

• DO NOT Run on Production: redis-benchmark can saturate the CPU and network, severely impacting the performance of your live application. Always run it on a dedicated test server with identical hardware and network configuration to production.
• Simulate Real Workloads: The default benchmark is generic. Use the -t, -d, and -c options to mimic your application’s specific command patterns, data sizes, and concurrency levels for a more accurate assessment.
• Test Pipelining: Use the -P option to see the significant performance boost from pipelining. For example, -P 16 will send commands in batches of 16.
• It’s a Baseline: Benchmark results are a baseline, not an absolute guarantee of real-world performance. Application logic, network fluctuations, and other factors will have an impact. Use it to compare configurations, hardware, or Redis versions.