Out of sight, out of mind -- that's the usual attitude toward hard disk storage. Until something goes wrong. Then it's "disarranging" time as described in the Stones' classic, "19th Nervous Breakdown."

While content creators often focus on storage performance for functions such as Photoshop scratch disks and video streaming, the reliability of your hard drive is just as critical to your success. We just assume that our hard drives are reliable. And mostly they are.

So my interest was piqued by a recent brouhaha online about changes in measurement of hard drive reliability. The postings in forums as well as in several news stories appeared to signal a fundamental shift in the standards for drives.

Taking Care of Business?
The flap concerned IBM's Deskstar 120GXP series of drives. Released last year, the 7,200-rpm drives come in capacities of 40, 80 and 120 GB (corresponding to 1, 2 and 3 platters, respectively).

Posts on hardware review and discussion sites such as Storage Review pointed out that the 120's warranty seemed to state that the drives should only be powered on for 333 hours per month, or about 10 hours a day.

As might be expected, readers expressed outrage over the perceived fine print and the hard drives' short time for usage. After all, we're used to running our hard drives (and the systems attached to them) for many hours a day, or days and weeks at a time. And besides, who wants to track the usage patterns of an individual drive? On or off, a hard drive should just work.

Could this be the beginning of some industry-wide performance bifurcation for primary storage? Would hard drive manufacturers take a page from the sales books of the cellular phone industry, making some mechanisms that you can run all day long, and others that are good only on the weekend?

That would be a nightmare.

Time Is On My Side
According to storage industry analyst Jim Porter, president of DISK/TREND, the whole megilla was "No big deal." The problem with the 120GXP had little to do with drive technology and everything with the process of crafting a warranty.

Porter said the genesis of the issue was the wish of computer manufacturers to get away from mean time between failure (MTBF), the tried-and-not-so-true statistical method for figuring hardware reliability. Drives that ship with desktop machines have a MTBF of 300,000 to 500,000 hours; very-high-performance drives designed for the server market (the ones with SCSI or FC-AL interfaces and spindle speeds of 10K or higher) can have an MTBF up to 1 million hours.

Of course, no company tests its drive models for 38, 70 or 114 years before releasing them into the market -- "nobody would be around to check," Porter joked. Instead, the MTBF figure is derived from testing a group of drives for a short time and then projecting the data.

As you can see in this MTBF FAQ, the world of statistics can be mysterious: The more drives you own, the lower the MTBF. So if you own 100 of the 114-year MTBF drives, you can expect at least one to die during the span of a year.

"Some desktop OEMs wrote drive specs asking drive vendors to guarantee 330 power-on hours per month for their 1-year, 2-year, 3-year, or 5-year warranties. So the drive vendors complied," said Mike Mihalik, a Portland-based consultant and the former vice president of engineering at LaCie Ltd., adding that IBM could have done a better job when writing the specification. "Unfortunately, taken out of context, it looks like you should only use the drives for 330 hours per month, or you void the warranty. You can see how all those 24/7 server users might get concerned."