FIO Storage I/O Benchmark

Tests storage I/O performance by running FIO benchmarks across multiple VMs in parallel.

Use Case: Measure storage IOPS, bandwidth, and latency under various workload patterns.

Basic Usage

virtbench CLI

# Run FIO benchmark on 10 VMs
virtbench fio \
  --start 1 --end 10 \
  --storage-class YOUR-STORAGE-CLASS \
  --save-results

# Custom FIO parameters
virtbench fio \
  --start 1 --end 20 \
  --storage-class YOUR-STORAGE-CLASS \
  --fio-runtime 300 \
  --fio-rw randrw \
  --fio-bs 4k \
  --fio-iodepth 64 \
  --save-results \
  --cleanup

Python Script

cd fio-benchmark

python3 measure-fio-performance.py \
  --start 1 --end 20 \
  --storage-class YOUR-STORAGE-CLASS \
  --fio-runtime 300 \
  --fio-rw randrw \
  --fio-bs 4k \
  --fio-iodepth 64 \
  --fio-numjobs 4 \
  --save-results \
  --results-dir ../results/my-test \
  --cleanup

Configuration Options

FIO Parameters

Option	Default	Description
--start	(required)	Starting namespace index
--end	(required)	Ending namespace index
--storage-class	(required)	Storage class name
--namespace-prefix	fio-benchmark	Namespace prefix (creates fio-benchmark-1, etc.)
--fio-runtime	300	FIO test duration in seconds
--fio-rw	randwrite	I/O pattern: read, write, randread, randwrite, randrw, rw
--fio-bs	4k	Block size (e.g., 4k, 8k, 64k, 1M)
--fio-iodepth	64	I/O queue depth
--fio-numjobs	4	Number of parallel FIO jobs per VM
--fio-size	10G	Test file size per job

Result Collection Options

Option	Default	Description
--collect-retries	8	Max retries for collecting results from VMs
--collect-retry-delay	20	Delay (seconds) between collection retries
--collect-concurrency	5	Max concurrent result collections

Output Options

Option	Default	Description
--save-results	false	Save results to JSON/CSV files
--results-dir	results	Base directory for results
--cleanup	false	Delete VMs and namespaces after test

Disk Space Requirements

FIO creates test files based on --fio-size and --fio-numjobs:

Setting	Calculation	Example
Test files per VM	fio-size × fio-numjobs	10G × 4 = 40GB
Scratch disk size	Must be > test files	50GB (in VM template)

Data written during test depends on achieved IOPS:

Data Written = Write IOPS × Block Size × Runtime

Example: 21,000 IOPS × 4KB × 60s ≈ 5GB written

Workload Examples

# Sequential read (max bandwidth)
virtbench fio --start 1 --end 10 --storage-class YOUR-SC \
  --fio-rw read --fio-bs 1M --fio-iodepth 32 --save-results

# Random write (database workload)
virtbench fio --start 1 --end 10 --storage-class YOUR-SC \
  --fio-rw randwrite --fio-bs 4k --fio-iodepth 64 --save-results

# Mixed read/write 50/50 (general workload)
virtbench fio --start 1 --end 10 --storage-class YOUR-SC \
  --fio-rw randrw --fio-bs 4k --fio-iodepth 64 --save-results

Results

Results include per-VM and aggregated metrics:

• IOPS: Read/Write operations per second
• Bandwidth: Read/Write throughput in MiB/s
• Latency: Read/Write latency in milliseconds

Output Structure

results/{timestamp}_fio_benchmark_{start}-{end}/
├── aggregated_results.json    # Summary across all VMs
├── aggregated_results.csv     # CSV format
└── per-vm-results/            # Individual VM results
    ├── fio-benchmark-1/
    │   └── fio_raw.json
    └── ...

View in Dashboard

cd dashboard
python3 generate_dashboard.py --base-dir ../results
open output/dashboard.html

The dashboard displays:

• KPI summary cards (Total IOPS, Bandwidth, Latency)
• IOPS + Latency dual-axis chart
• Per-VM breakdown with clickable namespaces