Test the IO performance of hard-disks/disks using Linux

Measure writing performance. This test can also be used with CIFS and NFS and is relatively objective:

dd if=/dev/zero of=temp.bin bs=1M count=1024 conv=fdatasync,notrunc

The parameter fdatasync only ends the dd command when the data has been completely synced. Instead, you could also use oflag=dsync which also takes caches into account and waits until they have been written.

To test without a buffer, we recommend the parameter oflag=direct using dd. In addition, the write cache of the disc should deactivated using:

hdparm -W0 /dev/sda

It can be re-activated using:

hdparm -W1 /dev/sda

The reading performance can be obtained with the command

time dd if=/tmp/test.bin of=/dev/null bs=4k

bs should be matched to the source drive.

Before doing read tests, it is recommended to flush the read cache:

echo 3 | sudo tee /proc/sys/vm/drop_caches

In all tests, where caches play a role, the RAM’s performance is of course also decisive!

Determine data throughput

This point complements the information above, but comes from a different source:

if=/dev/zero of=/tmp/test1.img bs=1G count=1 oflag=dsync

Determine latency

This point complements the information above, but comes from a different source:

dd if=/dev/zero of=/tmp/test2.img bs=512 count=1000 oflag=dsync

To test the performance of the cache, hdparm can be used.

hdparm -Tt /dev/sda

Using -t the performance of buffered read accesses is determined. This test primarily determines the performance between disk, kernel and chipset including system caches. The optional parameter –direct bypasses system caches and shows the direct data throughput between disk, kernel and chipset.

With -T the read cache is tested, this test does not make hard disk access because it only reads from the Linux buffer cache.