Yes, Virginia, RAID Drives Are Different

While some other members of the storage industry's chattering class enjoy the process of haranguing array vendors for their seemingly outrageous disk drive markups, I've started recommending that users seeking cost-effective online storage look at BYOD (bring your own drive) storage systems. There are many choices available, from prosumer Data Robotics offerings and SOHO network-attached storage (NAS) boxes from NetGear and QNAP Systems to disk arrays with enterprise aspirations provided by Promise Technology and Infortrend. However, care must be taken when choosing the drives for a BYOD array.

The truth is that the drive vendors haven't really tried to educate anyone but their OEM customers about the subtle differences among the various models of drives they offer at the same capacity point with the same interface. Even friends whom most would consider storage experts haven't spent a lot of time trying to figure out what makes a Western Digital RAID Edition drive different from a Caviar system.

Since most professional storage analysts and advocates are also significant storage consumers--with either a home lab or just every DVD they ever wanted available through Apple TV or Boxee around the house--many have ended up with a Data Robotics Drobos system of one kind or another. Data Robotics has even seeded units to a few of us, so I have a Drobo Elite in the lab courtesy of those nice folks.  

So we see tweets and blog posts like "Which Hard Disk Drives Should You Use in a Drobo?" from fellow Network Computing blogger Stephen Foskett. The responses to the blog describe users' vendor and model experiences, which are, of course, based on sample sizes too small to be really meaningful, but not the technology.

So what makes a "nearline" drive different than a desktop drive with the same capacity? I set out to answer that question a few months ago and sent e-mails to my contacts at the usual drive vendors. None had any real answers. The best response was, "No one that knows the answer to that question is press-trained."Nosy geek that I am, I investigated further. I discovered that while the mechanical head disk assembly (HDA) hardware for a 1TByte desktop and RAID drive may be very similar, there are some significant differences in the firmware that make RAID drives worth the extra money for BYOD arrays and a bad idea for desktops.

The main differences are in how the drives deal with errors and rotational vibration. Since RAID drives are used in groups, the torque from the adjacent drive's spindle motors and positioners can move the heads of a drive off track. RAID drives expect these effects and perform better when exposed to higher levels of rotational vibration.

When an error occurs, desktop drives are programmed to conduct a lengthy retry cycle that can take up to 30 seconds trying to read a data block. Since a desktop drive holds the only copy of a data item, extensive retries--including moving the heads off track and returning--are appropriate. The problem is that RAID controllers will see the drive go offline for a few seconds while it conducts its retries and mark the failed drive as dead.

RAID drives have what Western Digital calls TLER, or Time Limited Error Recovery. Basically, the drives stop retrying and return an error so the RAID controller can recover the data from the mirror or parity disk. Faster error returns mean faster performance on read errors and are a good thing for a RAIDset.

However, some enthusiasts are convinced that "enterprise" drives have better bearings or other hardware, and use them as stand-alone drives or in RAID 0 sets. Not only are they wasting money, but TLER means they're more, not less, likely to lose data. There may be other differences that the vendors haven't told me about, and you can buy 2TByte drives with SAS interfaces to support dual porting to redundant controllers, but error recovery and vibration resistance are enough to get me to cough up $249 for a 2TByte Constellation ES rather than $169 for a Barracuda XT.