Reliability
Nearline vs. Desktop: Deploying Value in the Data CenterSpec sheets do hint at a few points in the enterprise drive’s favor. The first of these is arguably reliability and drive longevity. For many years, mean time between failures (MTBF) was the key metric on this front, with enterprise drives featuring a 1.2 million-hour estimate and consumer drives typically offering a spec half to one-quarter of that. More recently, the industry has added alternative reliability metrics that even a 300,000-hour (34.25-year) MTBF seems difficult to grasp.
The MTBF doesn’t imply that any given drive will run for 34 years. Rather, if a large drive batch were run under optimal conditions and their failures plotted as a bell curve, the peak of the curve would be at the 300,000-hour mark. But nobody cares about the curve’s mid-point. What matters is the left side of the bell curve, where failures start to become statistically significant. Because of this, some vendors started conveying reliability in terms of an annualized failure rate (AFR). MTBF and AFR are mathematically two sides of the same coin.
Some vendors still list contact start/stop cycles, which refers to when the drive heads vacillate from flying above the disk platters as they spin to resting on the platters’ “landing zones” when the platters are at rest. Today, though, most drives use a head load/unload cycle rather than landing zones, and load/unload cycles entail much larger specification numbers. For example, the Seagate Desktop HDD (formerly the Barracuda® drive) went from specifying a minimum of 50,000 contact start/stop cycles with the 500GB model to specifying at least 300,000 load/unload cycles with the 1TB. The otherwise very similar drive moved to a new head resting design, and the specification changed. It’s important to notice which spec you’re dealing with.
As a reference point for how these specs overlap, consider the Barracuda 7200.12 drive, which noted several specs in its documentation. The drive listed an MTBF of 750,000 hours and an AFR of 0.32% at an ambient temperature of 25˚C. The drive specified 10,000 start/stop cycles per year and 50,000 contact start/stop cycles. All of these were figured with the assumption of 2,400 power-on-hours per year. This last point is important because it conveys a load level typical of a consumer-class user: eight hours per day, five days per week for five years.

Today, the Desktop HDD specifies 2,400 power-on hours and 300,000 load/unload cycles. Dig into the manual, though, and you’ll find another specification: an average workload rate of up to 55TB per year. The drive does not specify MTBF or AFR numbers.
In contrast, the Constellation ES drive notes a 1.4 million-hour MTBF and 0.63% AFR based on 24x7 operation. Similarly, the drive’s power-on hours are rated at 8,760 hours with an average annual workload of up to 550TB per year – ten times that of the Desktop HDD. The drive is also rated for 600,000 load/unload cycles.
Why does all of this matter? Because desktop drives are very unlikely to survive for five years under enterprise conditions. No vendor has yet published a study detailing a desktop vs. enterprise drive analysis under, say, three-year data center conditions, but rather than considering the problem in terms of hours or load/unload cycles, one might assess the problem in terms of total work done over the drive’s life. This is why the terabytes-per-year spec is so important. If a desktop drive is rated to convey 55TB per year for five years, reliability issues will set in if you exceed that annual load. Failure rates will climb, necessitating the purchase of additional replacement drives and increasingly placing more strain on IT to handle those replacements.
Meanwhile, the enterprise drive suffers from no such degradation. The desktop drive is warranted for only two years while the Constellation ES drive is covered for five years. In other words, given our very loose scenario here, the data center customer might expect 110TB of service from the drive before falling out of warranty while the enterprise drive could exceed 2750TB. It follows that a data center manager might go through many desktop drives before equaling the total work done by a single enterprise drive. Given the pricing we mentioned at this article’s start, it would only take one such replacement for the desktop option to reach price parity with the enterprise alternative. In fact, as stated above, the enterprise drive would be ahead at this point owing to the maintenance costs associated with IT replacing the failed drive. It all boils down to total cost of ownership.
