A high-performance drive benchmarking utility designed to identify storage bottlenecks and performance outliers across multiple drives in distributed storage systems.
dperf is an enterprise-grade storage performance measurement tool that helps system administrators and DevOps teams quickly identify slow or failing drives in production environments. By performing parallel I/O operations across multiple drives and presenting results in a clear, sorted format, dperf enables rapid troubleshooting and capacity planning for storage infrastructure.
- Parallel Performance Testing: Simultaneously benchmark multiple drives to identify performance outliers
- Direct I/O Operations: Uses O_DIRECT for accurate drive performance measurement, bypassing OS caches
- Sorted Results: Automatically ranks drives by throughput, showing fastest drives first
- Flexible Workloads: Configurable block sizes, file sizes, and concurrency levels
- Production-Ready: Minimal resource footprint with automatic cleanup
- Enterprise Support: Multi-architecture Linux support (amd64, arm64, ppc64le, s390x)
- Kubernetes Native: Easily deploy as Jobs or DaemonSets for cluster-wide storage validation
For Enterprise Operations:
- Proactive Monitoring: Identify failing drives before they impact production workloads
- Hardware Validation: Verify new storage hardware meets performance SLAs
- Capacity Planning: Establish performance baselines for storage infrastructure
- Compliance: Document storage performance for audit and compliance requirements
- Cost Optimization: Identify underperforming drives that should be replaced
For Developers:
- CI/CD Integration: Automated storage performance validation in deployment pipelines
- Troubleshooting: Quick diagnosis of I/O performance issues
- Benchmark Comparisons: Compare different storage configurations and technologies
- Minimal Dependencies: Single binary with no external requirements
# Download the latest binary for Linux amd64
wget https://github.com/minio/dperf/releases/latest/download/dperf-linux-amd64 -O dperf
chmod +x dperf
# Benchmark a single drive
./dperf /mnt/drive1
# Benchmark multiple drives in parallel
./dperf /mnt/drive{1..6}Download the appropriate binary for your architecture:
| OS | Architecture | Binary |
|---|---|---|
| Linux | amd64 | linux-amd64 |
| Linux | arm64 | linux-arm64 |
| Linux | ppc64le | linux-ppc64le |
| Linux | s390x | linux-s390x |
# Example: Install on Linux amd64
wget https://github.com/minio/dperf/releases/latest/download/dperf-linux-amd64
sudo install -m 755 dperf-linux-amd64 /usr/local/bin/dperfRequires Go 1.24 or later. Install Go if not already available.
# Install directly from source
go install github.com/minio/dperf@latest
# Or clone and build
git clone https://github.com/minio/dperf.git
cd dperf
make build
sudo make install# Benchmark a single drive
dperf /mnt/drive1
# Benchmark multiple drives in parallel (default mode)
dperf /mnt/drive{1..6}
# Run benchmarks sequentially (one drive at a time)
dperf --serial /mnt/drive{1..6}
# Verbose output showing individual drive statistics
dperf -v /mnt/drive{1..6}
# Write-only benchmark (skip read tests)
dperf --write-only /mnt/drive{1..6}
# Custom block size and file size
dperf -b 8MiB -f 5GiB /mnt/drive{1..6}
# High concurrency test
dperf -i 16 /mnt/drive{1..6}Flags:
-b, --blocksize string Read/write block size (default "4MiB")
Must be >= 4KiB and a multiple of 4KiB
-f, --filesize string Amount of data to read/write per drive (default "1GiB")
Must be >= 4KiB and a multiple of 4KiB
-i, --ioperdrive int Number of concurrent I/O operations per drive (default 4)
Higher values increase parallelism
--serial Run tests sequentially instead of in parallel
Useful for isolating drive-specific issues
--write-only Run write tests only, skip read tests
Faster benchmarking when only write performance matters
-v, --verbose Show per-drive statistics in addition to aggregate totals
-h, --help Display help information
--version Show version information
$ dperf /mnt/drive{1..4}
┌────────────────┬──────────────┐
│ TotalWRITE │ TotalREAD │
├────────────────┼──────────────┤
│ 4.2 GiB/s │ 4.5 GiB/s │
└────────────────┴──────────────┘With verbose output (-v):
$ dperf -v /mnt/drive{1..4}
┌──────────────┬──────────────┬──────────────┬────┐
│ PATH │ WRITE │ READ │ │
├──────────────┼──────────────┼──────────────┼────┤
│ /mnt/drive1 │ 1.1 GiB/s │ 1.2 GiB/s │ ✓ │
│ /mnt/drive2 │ 1.0 GiB/s │ 1.1 GiB/s │ ✓ │
│ /mnt/drive3 │ 1.1 GiB/s │ 1.2 GiB/s │ ✓ │
│ /mnt/drive4 │ 1.0 GiB/s │ 1.0 GiB/s │ ✓ │
└──────────────┴──────────────┴──────────────┴────┘
┌────────────────┬──────────────┐
│ TotalWRITE │ TotalREAD │
├────────────────┼──────────────┤
│ 4.2 GiB/s │ 4.5 GiB/s │
└────────────────┴──────────────┘Deploy dperf as a Kubernetes CronJob to regularly validate storage performance:
# Run daily storage health checks
kubectl apply -f - <<EOF
apiVersion: batch/v1
kind: CronJob
metadata:
name: storage-health-check
spec:
schedule: "0 2 * * *"
jobTemplate:
spec:
template:
spec:
containers:
- name: dperf
image: quay.io/minio/dperf:latest
args: ["/data1", "/data2", "/data3", "/data4"]
volumeMounts:
- name: storage
mountPath: /data
restartPolicy: OnFailure
EOFBefore adding new storage nodes to production, validate performance meets requirements:
# Define performance SLA
MIN_WRITE_THROUGHPUT="800MiB/s"
MIN_READ_THROUGHPUT="900MiB/s"
# Run benchmark and validate
dperf /mnt/new-drive{1..8} > results.txt
# Parse results and compare against SLAQuickly identify the slowest drives in a storage cluster:
# Sorted output shows slowest drives at the bottom
dperf /mnt/drive{1..100} | tee drive-performance.logTest storage performance for Kubernetes PersistentVolumes before application deployment:
apiVersion: batch/v1
kind: Job
metadata:
name: pv-performance-test
spec:
template:
spec:
restartPolicy: Never
containers:
- name: dperf
image: quay.io/minio/dperf:latest
args: ["-v", "/data"]
volumeMounts:
- name: test-volume
mountPath: /data
volumes:
- name: test-volume
persistentVolumeClaim:
claimName: my-pvc- Operating System: Linux (uses Linux-specific direct I/O features)
- Kernel: Linux kernel with O_DIRECT support
- File System: Any file system supporting direct I/O (ext4, xfs, etc.)
- Permissions: Write access to target directories
- Disk Space: Temporary space equal to
filesize × ioperdriveper drive
dperf performs accurate drive performance measurements using several key techniques:
- Direct I/O (O_DIRECT): Bypasses operating system page cache for accurate hardware measurements
- Page-Aligned Buffers: Uses 4KiB-aligned memory buffers required for direct I/O operations
- Parallel Testing: Launches concurrent I/O operations per drive to measure maximum throughput
- Fsync/Fdatasync: Ensures data is written to physical media, not just cached
- Sequential Hints: Uses
fadvise(FADV_SEQUENTIAL)for optimized read patterns
The benchmark process:
- Creates temporary test files in each target directory
- Writes data using multiple concurrent threads (default: 4 per drive)
- Reads data back using the same concurrency level
- Calculates throughput (bytes/second) for both operations
- Cleans up all temporary files automatically
# Ensure write permissions on target directories
sudo dperf /mnt/restricted-drive
# Or change directory permissions
sudo chmod 755 /mnt/restricted-driveBlock size must be at least 4KiB and a multiple of 4KiB due to direct I/O alignment requirements:
# Invalid
dperf -b 1024 /mnt/drive1 # Too small
# Valid
dperf -b 4KiB /mnt/drive1 # Minimum size
dperf -b 1MiB /mnt/drive1 # Common size
dperf -b 8MiB /mnt/drive1 # Larger blocks- Check Drive Health: Use
smartctlto verify drive health - Check CPU Throttling: Ensure CPU is not throttled during tests
- Check I/O Scheduler: Verify appropriate I/O scheduler is configured
- Increase Concurrency: Try higher
-ivalues for better parallelism - Check File System: Some file systems perform better than others
Contributions are welcome! Please feel free to submit a Pull Request. For major changes, please open an issue first to discuss what you would like to change.
dperf is released under the GNU Affero General Public License v3.0. See LICENSE for details.
- Issues: GitHub Issues
- Enterprise Support: Contact MinIO Sales