Hands On: Which Raid is right for you?

The unedited video files and the final rendered edit would be stored on the Raid array to test both its read and write performance in a real-life environment. Such real-world tests are crucial as a pure benchmark may not provide an accurate account of how you’ll actually use the hardware.

The results
For comparison I started by testing just one of the 500GB disks connected directly to the motherboard’s controller. HD Tach gave average read and write speeds of 58.8 and 63.8Mbytes/sec respectively and a burst rate of 227.5Mbytes/sec; the Premiere render took 195 seconds.

I then connected two of the 500GB disks as a Raid 1 array (giving a 500GB total capacity) using a 64KB stripe size and 512KB sector size. HD Tach gave average read and write speeds of 66.7 and 34.3Mbytes/sec respectively and a burst rate of 140.8Mbytes/sec, while the Premiere render took 181 seconds. So while HD Tach reported slightly faster read and much slower write speeds, the final Premiere job was still quicker overall.

Next, the two disks were set up as a Raid 0 array (1TB total capacity), again using the default 64KB and 512KB stripe and sector sizes. This time HD Tach reported average read and write speeds of 110.8 and 97.2Mbytes/sec respectively and a 196.7Mbytes/sec burst rate, while the Premiere render took 168 seconds. Both HD Tach scores were clearly much faster, which is reflected in the quicker render.

I then connected a third disk and built a new Raid 0 array (1.5TB total capacity), again with the defaults. HD Tach reported average read and write speeds of 110.9 and 91.9Mbytes/sec respectively and a burst rate of 246.8Mbytes/sec, while the Premiere render took 156 seconds. The average read score may have been much the same and the write a little slower, but perhaps influenced by the much faster burst rate, the render was the quickest yet.

I then switched to Raid 5 using the three disks (1TB total capacity), but also reduced the stripe size to 32KB. HD Tach gave average read and write speeds of 66.6 and 69.2Mbytes/sec respectively and a 196.2Mbytes/sec burst rate, while the Premiere render took 240 seconds. HD Tach’s figures may have been a little quicker than for the single disk, but the Premiere render was the slowest yet.

In the next test I increased stripe size to the default 64KB. HD Tach reported average read and write speeds of 72 and 80.9Mbytes/sec respectively and a burst rate of 239.6Mbytes/sec, while the Premiere render took 218 seconds – a big improvement to all the scores, just by adjusting the stripe size.

Next, the biggest stripe size of 128KB saw HD Tach report average read and write speeds of 68.2 and 81.7Mbytes/sec respectively and a burst rate of 238.2Mbytes/sec, while the Premiere render took 210 seconds. A slight adjustment across the HD Tach results, but again a quicker render, if not as quick as the original single disk.

I then repeated the Raid 5 tests using the three different stripe sizes at each of the controller’s sector options of 1,024, 2,048 and 4,096KB; the setting used above was the default 512KB. The average read and write times remained pretty similar, although larger sectors improved the burst rates. The Premiere renders were within a few seconds of the default scores, though, so nothing to get too excited about for this particular application.

Big variations
The results prove there are significant differences in the performance of each Raid level, and subtler, but measurable differences between stripe sizes. Revealingly, though, the scores measured by a pure benchmark such as HD Tach may not be proportionately or even accurately reflected in a real-life environment.