How to benchmark disk speed in Linux

Disk benchmarks in Linux help confirm whether slow backups, package installs, database jobs, or VM storage are waiting on the disk path. A short read and write test gives an operator a baseline before moving data, replacing hardware, or tuning applications.

Use lsblk to map the whole disk, partitions, and mount points before running a test. hdparm measures read timing from the block device, while dd writes a temporary file through the mounted filesystem so the write number reflects the path used by normal files.

Run benchmarks on a quiet host and keep the device and file path consistent when comparing results. The hdparm timing options read from a whole disk such as /dev/sda or /dev/nvme0n1, while the dd write target must be a regular file such as /var/tmp/disk-benchmark.img rather than a block device.

1. Map the disk and mount point that will be tested.

   $ lsblk --output NAME,PATH,SIZE,TYPE,MOUNTPOINTS
   NAME        PATH           SIZE TYPE MOUNTPOINTS
   sda         /dev/sda       100G disk
   ├─sda1      /dev/sda1        1G part /boot/efi
   └─sda2      /dev/sda2       99G part /
   nvme0n1     /dev/nvme0n1   500G disk
   └─nvme0n1p1 /dev/nvme0n1p1 500G part /data

   Use the whole-disk path, such as /dev/sda or /dev/nvme0n1, for hdparm. Use a directory on the mounted filesystem, such as /var/tmp or /data, for the dd write file.

2. Measure cached and buffered read timing on the selected disk.

   $ sudo hdparm -Tt /dev/sda
   
   /dev/sda:
    Timing cached reads:   36258 MB in  1.99 seconds = 18224.37 MB/sec
    Timing buffered disk reads: 4824 MB in  3.00 seconds = 1607.72 MB/sec

   Timing buffered disk reads is the device read number to compare. Timing cached reads mostly measures memory and kernel page-cache speed.

3. Measure direct read timing when a page-cache-free device baseline is needed.

   $ sudo hdparm -t --direct /dev/sda
   
   /dev/sda:
    Timing O_DIRECT disk reads: 9174 MB in  3.00 seconds = 3057.77 MB/sec

   --direct makes hdparm use O_DIRECT for the -t timing test, bypassing the page cache.

   Disk benchmarks increase live I/O. Avoid running them during busy production periods or on a block device selected by guesswork.

4. Write a temporary 1 GiB test file on the mounted filesystem.

   $ dd if=/dev/zero of=/var/tmp/disk-benchmark.img bs=64M count=16 oflag=direct status=progress
   16+0 records in
   16+0 records out
   1073741824 bytes (1.1 GB, 1.0 GiB) copied, 0.675606 s, 1.6 GB/s

   The of= target must be a regular file on a mounted filesystem. Pointing of= at a device such as /dev/sda overwrites disk data.

   oflag=direct bypasses the page cache for the file write. If the filesystem rejects direct I/O with Invalid argument, rerun the same test with conv=fdatasync so dd flushes the file before it exits.
   Related: How to check dd progress in Linux
   Related: How to show disk activity on Linux

5. Remove the temporary benchmark file.

   $ rm -f /var/tmp/disk-benchmark.img

   Leaving the file in place consumes real disk space and can skew later capacity checks.

6. Repeat the same read and write tests when comparing disks or before-and-after storage changes.

   Keep the whole-disk device, mounted test directory, block size, total write size, and system load as similar as possible so the hdparm and dd numbers remain comparable.