Raid 1 Vs Raid 5 Speed Test Speed Test

I will be using two servers that are identical in every way, except for the RAID configuration they are using. The first server has been configured with a single large volume consisting of one disk, the second server is exactly the same aside from having five disks configured as a RAID 5 array.

The drives used in this test are WD1002FBYS 1TB SATA3 7200rpm drives, with 128MB cache and two heads each. They were utilized in a mirrored configuration on the first server, as well as RAID 5 on the second server.

It is important to note that this review is not an indication of which configuration will work best for a given user, but rather an objective comparison of the two configurations for speed.

The network utilized is a 100mbit/s Ethernet connection. This was used to connect directly to both servers whilst running multiple tests from a single desktop computer, using IxChariot software in order to average the results over several runs and eliminate the possibility of anomalies.

IP address and login details for the servers were recorded in order to replicate identical load tests on both configurations, all other system settings are left at default values (no overclocking was used).

The first test is a simple file copy from one server to another, using only one core of the desktop computer's processor. In this test the data being copied is a single 1GB text file, using Windows Explorer to drag and drop the additional files.

The second test attempts to copy an entire 1TB volume, in this case leaving only 7% free disk usage on the source server (the target had 99%).

The third test uses real-world performance data from an Iometer test performed on the server's volume. This data is then duplicated to the second server and read back, in this way mimicking real-world file access by clients.

The fourth test again uses a real-world performance benchmark but this time measures write performance rather than read speeds. This tests how quickly files can be written to the volume, rather than read.

In several cases RAID 5 is almost 50% faster than mirroring, and more than 15% faster in all tests aside from streaming real data (which could be expected to perform worse due to the greater number of drives). It is also interesting to note that performance during reads was only marginally less than for writing, which is not a result I expected.

Conclusion

RAID 5 definitely appears to have an advantage over RAID 1 in terms of throughput and generalized performance. Of course, the inclusion of a hot spare may make up for this in some circumstances that involve hard drive failures.