Reason #5: Data Integrity
There’s More To An Enterprise Drive
What's this
Enterprise-class hard drives consume more energy than many of their desktop counterparts, especially so-called “green” drives. However, the wattage number you see on the spec sheets can be misleading.
Consider the Seagate Barracuda 3TB Barracuda (ST3000DM001) desktop drive. This consumer-class SATA drive uses three platters, idles at 5.40W, and consumes 8.00W under operating load. The 3TB Constellation ES.2 (ST33000650SS) idles at 7.40W and averages 11.33W under load. On the surface, this looks like much greater power efficiency for the Barracuda. But keep in mind that companies don’t care about spot wattage measurements. They care about total power consumed over time.
As noted earlier, Turner’s experiments revealed that RV tolerance measures can yield very significant performance increases. Data centers measure performance in I/O operations per second, and many organizations run jobs of a set size daily. Say that it takes a consumer drive four hours to complete a job at 8W. The enterprise drive will consume 11W performing the same work, but, because of the many reasons described above, it might complete the job in only two hours. The desktop drive will consume 0.032 kWh while the enterprise drive chews through only 0.022 kWh—a roughly 45% power savings for the enterprise drive on that job. Magnify that difference across hundreds or thousands of drives running around the clock and the cost savings on energy become obvious.
Some organizations are finding that migrating to 2.5” hard drives, such as Seagate’s 1TB Constellation, provide enterprise-class reliability benefits while dropping the power per drive down to 3.84W average idle and 6.35W average load. In many cases high-density cases, this can yield lower total power consumption for the same aggregate drive capacity.
Alternatively, all late-model Seagate enterprise drives use a new set of technologies the company calls PowerChoice, which is based on the T10 technical committee standard 09-054 and T13 technical committee standard 452-2008. These involve a set of protocol-level commands that IT can script and configure to push drives into four possible power states based on different trigger conditions. The least impactful of these, Idle_A, leaves the platters rotating at full speed but disables some of the servo system and reduces power to the processor. At the deepest level, Standby_Z, heads are parked, the drive motor spins down, although the drive can still respond to “non-media” commands. As you can see, aggressive use of these modes can result in a power savings of up to 54%.
As with other computing systems in the data center, every watt saved at the storage component level means, give or take, another watt saved in cooling. By simply giving IT staff the tools to implement storage more effectively, Seagate is helping organizations save potentially thousands of dollars annually on their power consumption alone.
Reason #5: Data Integrity
Broadly speaking, data integrity refers to the ability of data stored on media to remain intact over time. Use, aging, stray energy particles, and other factors can lead to data becoming corrupt as it resides in a drive's storage medium. In the paper "An Analysis of Data Corruption in the Storage Stack," Lakshmi Bairavasundaram, et. al. examined 1.53 million nearline (SATA) and enterprise (Fibre Channel, which is based on SCSI commands rather than nearline's AT command set) hard drives over 41 months. The authors observed over 400,000 cases of data corruption. While some of these can be attributed to mechanical issues, the paper goes on to detail "silent data corruptions" and the role that enterprise-class drive design can play in data integrity.
Less well-known, however, is that current hard drives and controllers consist of hundreds-of-thousands of lines of low-level firmware code. This firmware code, along with higher-level system software, has the potential for harboring bugs that can cause a more insidious type of disk error – silent data corruption, where the data is silently corrupted with no indication from the drive that an error has occurred. ... We find that nearline (SATA) disks and their adapters develop checksum mismatches an order of magnitude more often than enterprise class (FC) disks. ... Using our probability representation, P(Xt ≥ 1) = 0.0086 for nearline disks, and P(Xt ≥ 1) = 0.00065 for enterprise class disks without any restriction on time.