Benchmarking Linux disk and SSD speed with hdparm, dd and fio

Since a drive is often the bottleneck for server performance, it’s worth being able to quickly measure its actual throughput. Three commonly available tools cover most needs, from a quick sanity check to a realistic sustained-load test.

Prerequisites

• Root/sudo access to run the reads/writes below.
• fio may need installing; hdparm and dd are pre-installed on almost every Linux distribution.

Steps

1. Quick read speed check with hdparm

hdparm -tT /dev/<device>

-t measures raw read speed from the device; -T measures read speed from cache. Replace /dev/<device> with the drive to test.

To test write speed:

hdparm -tT --write-sector /dev/<device>

Warning: this test can damage data on the drive. Only run it against a drive whose data you don’t need.

2. Simple sequential throughput with dd

dd if=/dev/zero of=/tmp/test bs=1G count=10 oflag=direct

Copies 10 blocks of 1 GB each from /dev/zero to /tmp/test. oflag=direct bypasses the page cache so the result reflects real device write speed rather than RAM buffering. Delete the test file afterward.

3. Realistic random I/O load with fio

apt install fio

fio --name=randwrite --ioengine=libaio --rw=randwrite --bs=4k --size=100G --numjobs=1 --runtime=120s --time_based --group_reporting

--rw=randwrite runs a random-write pattern (closer to real-world database/VM workloads than sequential dd); --bs sets the block size, --size the test file size, --numjobs the number of parallel jobs, and --runtime/--time_based bound the test to a fixed duration rather than a fixed amount of data.

Verify

Compare results across runs/drives using the same parameters (block size, size, job count) — raw numbers are only meaningful relative to a consistent test setup, not as absolute figures to compare against unrelated benchmarks online.

Troubleshooting

• hdparm numbers look too good to be true: you may be measuring cache (-T) rather than the device itself (-t) — check which flag produced the number.
• dd numbers are inflated: confirm oflag=direct is present; without it you’re partly measuring page-cache write-back speed, not the device.

Related

Sources

• Legacy wiki.js: Linux hard disk benchmarking tools — private source