Benchmark

Path: Left sidebar > Monitoring > Benchmark

When to Use:

• After hardware, topology, or performance-policy changes.

• Before onboarding sensitive workloads or validating a baseline.

Purpose:

This page shows how to run controlled disk, pool, and RBD benchmarks and compare the result with your expected baseline.

Steps:

1. Open Monitoring > Benchmark.

2. Select the benchmark tab that matches the layer you want to test.

3. Choose the target node, configure the run, and start the benchmark.

4. Review Benchmark History and compare the results with your baseline.

Expected Outcome:

• You get repeatable throughput, IOPS, and latency measurements for the selected layer.

What You See:

• Benchmark tabs, target-node selection, run forms, and Benchmark History for completed tests.

What This Screenshot Shows:

• The screenshots on this page show the disk, pool, and RBD benchmark forms used to launch and review test runs.

Actions in This Screen:

• Select a target node.

• Run disk, pool, or RBD benchmarks.

• Review completed benchmark history.

If this fails:

1. Confirm the target node and selected resource are healthy and reachable.

2. Retry during a lower-load window if the cluster is already busy.

3. Clean up old test data before repeating the run when required.

Benchmark Overview

Ceph Benchmark lets you run synthetic I/O tests against the cluster to measure raw performance of disks, pools, or RBD images. Use it after hardware changes or to validate SLA targets.

Warning

Benchmarks generate real I/O load. Run them during low-traffic periods to avoid impacting production workloads.

Benchmark Tabs

Tab	What It Tests	Tool Used
Disk Benchmark	Raw sequential and random performance of individual disks on a host	fio
Pool Benchmark	Ceph pool performance through direct RADOS operations	rados bench
RBD Benchmark	Performance of a specific RBD image simulating VM-like workload	rbd bench

Purpose:

• To validate performance after hardware, disk, or topology changes.

• To compare measured results against SLA targets.

• To isolate bottlenecks at disk, pool, or image layer.

When to Use:

• After node or disk replacement.

• After adding OSDs or changing placement/replication behavior.

• Before onboarding latency-sensitive workloads.

Steps:

1. Select the benchmark tab for the layer you want to test.

2. Click top-right Select target node... and choose the host.

3. Configure test fields and run the benchmark.

4. Review Benchmark History and compare with expected baseline.

Expected Outcome:

• You get repeatable baseline metrics for IOPS, throughput, and latency.

Select Target Node

All tabs require a target node from the top-right Select target node... action button before any run can start.

Note

If no node is selected, run actions remain blocked and the UI prompts node selection.

Benchmark History

Each tab has a Benchmark History area where completed runs appear. The empty state is normal before first execution.

Disk Benchmark Tab

Path: Monitoring > Benchmark > Disk (fio)

What This Screenshot Shows: Benchmark - Disk (fio) Tab (UI Reference; Values Depend On Your Environment).

Uses fio to test raw sequential and random disk performance on a selected host.

Purpose:

• To verify host-level disk capability for random and sequential workloads.

When to Use:

• Before assigning a disk to production workload.

• After replacing disk firmware or physical hardware.

How To Run A Disk Benchmark

1. Open Disk (fio) tab.

2. Click top-right Select target node... and choose the host.

3. Select Device to test (for example /dev/sdb).

4. Select Test Type (sequential or random, read or write).

5. Set Block Size (for example 4K for random tests, 1M for sequential tests).

6. Set Duration in seconds.

7. Click Run Disk Benchmark.

Expected Outcome:

• The test runs on the selected host/device for the configured duration.

• Results are written into Benchmark History after completion.

• You get host-disk baseline values for IOPS, throughput, and latency.

Tip

Use 4K block size for random IOPS validation and 1M block size for sequential throughput validation.

Disk Benchmark - Field Reference

Field	Description
Host	Select the host whose disks you want to test
Device	The specific block device (for example /dev/sdb)
Test Type	Sequential read/write or random read/write
Block Size	I/O block size (4K for random, 1M for sequential)
Duration	How long the test runs in seconds

Disk Benchmark - Result Metrics

• IOPS

• Throughput (MB/s)

• Average latency

• 99th-percentile latency

Pool Benchmark Tab

Path: Monitoring > Benchmark > Pool (rados)

What This Screenshot Shows: Benchmark - Pool (rados) Tab (UI Reference; Values Depend On Your Environment).

Uses rados bench to measure a Ceph pool by writing and reading objects through the RADOS protocol.

Purpose:

• To measure pool-level behavior including replication overhead and placement impact.

When to Use:

• Before onboarding high-throughput applications to a pool.

• After changing pool policy or data placement behavior.

How To Run A Pool Benchmark

1. Open Pool (rados) tab.

2. Click top-right Select target node... and choose the host.

3. Select target Pool.

4. Select Operation (Write, sequential read, or random read).

5. Set Concurrency.

6. Set Object Size.

7. Set Duration in seconds.

8. Click Run Pool Benchmark.

Expected Outcome:

• Test objects are written/read directly in the selected pool.

• Benchmark History shows measured throughput, IOPS, and latency.

• You get pool-level performance baselines through native RADOS path.

Note

Sequential reads require a prior write pass to populate test objects. Run a write test first.

Tip

Start with a write run, then execute sequential/random reads on the same pool to get a comparable read baseline.

Pool Benchmark - Field Reference

Field	Description
Pool	Target pool for the benchmark
Operation	Write, sequential read, or random read
Concurrency	Number of parallel I/O threads
Object Size	Size of each test object
Duration	Test duration in seconds

Pool Benchmark - Result Metrics

• IOPS

• Throughput

• Average latency

• Maximum latency

• Number of objects written

RBD Benchmark Tab

Path: Monitoring > Benchmark > Block (rbd)

What This Screenshot Shows: Benchmark - Block (rbd) Tab (UI Reference; Values Depend On Your Environment).

Uses rbd bench to measure performance of a specific RBD image and simulate VM workload behavior.

Purpose:

• To validate image-level block performance before production VM placement.

When to Use:

• Before assigning RBD images to sensitive VM workloads.

• After changing image-level settings or pool layout.

How To Run An RBD Benchmark

1. Open Block (rbd) tab.

2. Click top-right Select target node... and choose the host.

3. Select target Pool / Image.

4. Set I/O Type (Read or Write).

5. Set I/O Size (4K-4M).

6. Set I/O Threads.

7. Set Duration in seconds.

8. Click Run RBD Benchmark.

Expected Outcome:

• The benchmark generates block I/O against the selected image.

• Results appear in Benchmark History for throughput and latency review.

• You get image-specific baseline performance for VM planning.

Tip

Start with lower I/O Threads and increase gradually. Higher thread counts can increase throughput but can also increase latency.

RBD Benchmark - Field Reference

Field	Description
Pool / Image	The RBD image to benchmark
I/O Type	Read or write
I/O Size	Request size (4K-4M)
I/O Threads	Parallelism; higher threads can increase throughput and latency
Duration	Test duration in seconds

Interpreting Results

Metric	Guidance
IOPS (random 4K)	HDD: ~200 IOPS, SSD: ~50K+, NVMe: ~500K+
Throughput (sequential 1M)	HDD: ~200 MB/s, SSD/NVMe: 500MB/s-3GB/s
Latency (avg)	< 1ms excellent, 1-5ms acceptable, > 10ms investigate
p99 latency	Should be < 10x average; large outliers indicate disk or network issues

When Results Are Lower Than Expected:

• Check if competing workloads were active during test.

• Check Recovery Rate under Monitoring > Performance > OSD Performance.

• Check SMART status in Infrastructure > Hosts > Device Health.

• Account for replication overhead on write-heavy tests.

Cleanup

Pool and RBD benchmarks write test objects that must be cleaned up after benchmarking.

Purpose:

• To reclaim used space and avoid stale test artifacts.

When to Use:

• Immediately after benchmark validation is complete.

How To Clean Up:

1. Open Benchmark History for the run.

2. Click Clean Up for the selected benchmark result.

3. Confirm cleanup and verify test objects are removed.

Expected Outcome:

• Test data is removed from the selected pool and benchmark footprint is cleared.

Troubleshooting - Benchmark

Problem You See	Most Likely Cause	What To Do
Run action is blocked	No target node selected	Click top-right Select target node..., choose a host, then rerun.
Pool dropdown is empty	No pools available or load issue	Verify pools in Storage > Pools and refresh.
Sequential read gives invalid/low results	No prior write pass	Run write benchmark first, then rerun read.
RBD image not found	Wrong image name or wrong pool	Confirm image in Block Storage > Block Images.
Results far below expected	Competing workload or active recovery	Rerun in low-traffic window and verify recovery is complete.
Cleanup did not remove expected data	Wrong benchmark target selected	Recheck selected pool/image and run cleanup again.