Testing Disk Response Speed / Performance on a Linux Server

The disk subsystem on a Linux server seems to be slow. How can I benchmark the disk subsystem?

There are two tests that you can run to test the write speed and the read speed of a particular Linux partition.  You may also be interested in reading Understanding Disk I/O in Relation to Retain Performance.  The concepts would apply to Reload as well.

Write Test

dd if=/dev/zero bs=1024 count=1000000  of=/[path to volume you are testing]/disktest.txt

It will create a 1024 byte file called disktest.txt.  The byte size is variable, so you can increase it to a larger size just to make sure it isn't being done in cache (if you truly want to test the disk speed).

Example Output from a server manufactured in 2006 with local storage:

1000000+0 records in 1000000+0 records out 1024000000 bytes (1.0 GB) copied, 14.7302 seconds, 69.5 MB/s

Read Test

The read test must be performed after the write test above. Make sure to reference the file that the write test created.

dd if=/[path to volume you are testing]/disktest.txt bs=64k | dd of=/dev/null

Example Output from a server manufactured in 2006 with local storage:

15625+0 records in 15625+0 records out 1024000000 bytes (1.0 GB) copied, 4.83909 seconds, 212 MB/s 2000000+0 records in 2000000+0 records out 1024000000 bytes (1.0 GB) copied, 4.84089 seconds, 212 MB/s

Example Output from a server manufactured in 2007 with SAN attached storage:

15625+0 records in 15625+0 records out 1024000000 bytes (1.0 GB) copied, 6.71695 seconds, 152 MB/s 2000000+0 records in 2000000+0 records out 1024000000 bytes (1.0 GB) copied, 6.71715 seconds, 152 MB/s

Analyzing Your Server

If your server takes more than twice the amount of time for either the read or write tests than the above examples, then you need to consider how you configure or modify your server to get better speeds.  Reload and Retain are disk intensive applications.

Reload Implications If Reload backups are being done in a LAN enviornment (vs. a WAN) then the greatest bottleneck to speed is generally the speed of the disk on the Reload server. If the disk speed on your Reload server is slow, then decrease the BLOBS-Threads to a number that matches the speed of your disk. For example, rather than setting the BLOBS-Threads to 10, set them to 2. Also, do not try and run Concurrent Standard Backup Jobs, as that will only make running Reload backups unreliable and very slow. Also you should expect that creation of Portable Backups and TAR files is going to also be slow, as both of these operations are very disk intensive.

Retain Implications Retain archives jobs involve writing to disk heavily for these four areas:

• Archive files
• Indexes
• Database
• Logs

See the Retain Planning and Design Best Practices article for a more in-depth discussion on disk strategy.

This article was originally published in the GWAVA knowledgebase as article ID 1590.