Performance, Continued

Copy Tests

When benchmark drives or other forms of storage, I typically transfer between drives. Unfortunately, I don't have drives that stand up to the theoretical speeds that the array has shown in both hdparm and dd tests. Thusly, I'm forced to create a RAMdisk to test transfer speeds. A RAMdisk is a chunk of RAM converted to a mountable disk. Any data on the RAMdisk disappears after a reboot. A brief HOW-TO for RAMdisks can be found at Vanemery.com.

Let's look at the results of the RAMdisk's hdparm and dd: (Higher is better)

# hdparm -t /dev/ram0
/dev/ram0: 
Timing buffered disk reads:   2534MB in  3.01 seconds = 842 MB/sec

# dd if=/dev/zero of=/mnt/ram/test bs=1024 count=1000000
1.1 GB copied, 8.5874s seconds, 131 MB/s

With that much bandwidth, I should have no issues copy files to and from the drive. Since my RAMdisks are configured for 1GB in size, I decided to use a 700MB movie file for copying. After placing the movie on the RAID array, I copied it to the RAMdisk. Higher is better.

Afterwards, I copied it back.

Seek Times

Another important aspect of hard drives is random seek times. This determines how fast the disks finds a piece of data on the disk. The folks at LinuxInsight have a fantastic little script capable of checking this. It may not be the most accurate, but the results very much speak for the themselves when looking at a single drive against a RAID array. Lower is better.