When considering server performance, Raid is the most common way to achieve disk subsystem performance. When you are on a budget, the choices are very limited. But isn't that when your creativity hits the roof?
We needed cost-effective servers for virtualized environment. So we ran some hard drive benchmarks to see what we can extract.
We could use iometer or some other better mechanism of testing. But we simply wanted a quick and simple test.
So here are some numbers:
Normal Disk (No Raid)Disk Read:56 MB/secDisk Write:52 MB/secRaid-5Disk Read:138 MB/secDisk Write:117 MB/secRaid-10Disk Read:119 MB/secDisk Write:106 MB/sec
Note that on first look Raid-5 appears to offer the best performance on this system. Let's see if this holds true in a more realistic scenario. For further testing we started two threads, one reading and another writing. However they started one after another. Here are the numbers:
Raid-5 (write thread started first)Disk Read:46 MB/secDisk Write:78 MB/secRaid-5 (read thread started first)Disk Read:38.3 MB/secDisk Write:84.2 MB/secRaid-10 (write thread started first)Disk Read:54.7 MB/secDisk Write:108 MB/secRaid-10 (read thread started first)Disk Read:56 MB/secDisk Write:103 MB/sec
Now how about read performance with two threads:
Raid-5 – Concurrent ReadsThread 1:53.0 MB/secThread 2:59.7 MB/secRaid-10 – Concurrent ReadsThread 1:117 MB/secThread 2:111 MB/sec
Looking at these numbers, it is reasonable to conclude that Raid-10 offers a much more consistent performance with concurrent disk activity. Although these numbers do not consider all the situations however they consistently work for us.