Benchmarking

MinIO AIStor includes built-in performance testing tools to measure object operations, drive I/O, network bandwidth, site replication throughput, and client-to-server bandwidth.

For OS-level tuning recommendations, see Performance Tuning.

The mc support perf command runs performance tests against a MinIO AIStor deployment.

Run all performance tests:

mc support perf ALIAS

Run a specific test type:

mc support perf [object|drive|net|site-replication|client] ALIAS

Tests S3 PUT and GET throughput across the cluster:

mc support perf object ALIAS

Example output:

MinIO 2024.01.01, 4 servers, 16 drives, 64MiB objects, 32 threads
PUT: 2.5 GiB/s, 40 objs/s
GET: 3.2 GiB/s, 51 objs/s

Use --verbose to see per-server breakdown:

mc support perf object ALIAS --verbose

Expected results:

• Throughput scales linearly with server count.
• GET is typically 20-30% faster than PUT.
• Consistent per-server numbers indicate balanced load.

Tests raw disk I/O performance on each drive:

mc support perf drive ALIAS

Additional flags for the drive test:

Troubleshoot slow drives:

1. Check for failing drives: mc admin drive info ALIAS
2. Verify storage configuration.
3. Check for thermal throttling.
4. Review filesystem fragmentation.

Tests inter-node network bandwidth. Requires multiple servers in the cluster.

mc support perf net ALIAS

Results should approach physical network capacity. Troubleshoot network issues:

1. Check switch configuration.
2. Verify MTU settings (jumbo frames if supported).
3. Check for packet loss: ping -c 100 -s 1472 <node>
4. Review network interface errors with ethtool -S.

Tests cross-site replication bandwidth for multi-site deployments:

mc support perf site-replication ALIAS

Tests bandwidth from the mc client to the MinIO AIStor cluster:

mc support perf client ALIAS

Save results locally instead of uploading to SUBNET:

mc support perf ALIAS --airgap

Output is saved to ALIAS-perf_YYYYMMDDHHMMSS.zip.

Get results in JSON format for programmatic consumption:

mc support perf ALIAS --json

1. Initial deployment: Establish baseline performance.
2. Hardware changes: After adding or replacing drives or nodes.
3. Configuration changes: After tuning parameters.
4. Troubleshooting: When investigating performance issues.
5. Capacity planning: Before adding workload.

For consistent results:

1. Run during low-traffic periods.
2. Ensure cluster is healthy: mc admin info ALIAS
3. Run multiple iterations.
4. Test with representative object sizes.
5. Document system configuration.

A significant drop (greater than 10%) warrants investigation.

For detailed S3 benchmarking, use warp:

go install github.com/minio/warp@latest

warp mixed --host=server1:9000,server2:9000 \
  --access-key=minioadmin --secret-key=minioadmin

Warp provides:

• Detailed latency percentiles
• Mixed workload simulation
• Configurable object size distributions
• Multi-client coordination