Sign in with
Sign up | Sign in

The Factors Behind NLL

Moving Into the Lite
By
Brought to you by What's this

At first glance, the nearline lite category can look suspiciously similar to its desktop cousin. Because NLL does cost more per gigabyte, it’s important to have a clear understanding of the three factors at the core of NLL design: workload, temperature, and vibration.

Workload 

Traditionally, mean time between failures (MTBF) was considered one of the top measurements of drive reliability and longevity. While MTBF and its similar sibling, annualized failure rate (AFR), remain in widespread use, neither gives any sense of how much work a drive can be expected to perform in any given year. 

To understand the nature of workload differences between drives, think of pickup trucks. A 2013 Dodge RAM 3500 has a maximum payload capability of 6,580 pounds. This qualifies as “heavy-duty.” The 2013 Dodge RAM 1500 Sport offers a maximum payload of 1,374 pounds and qualifies as “light-duty.” Could the 1500 Sport physically carry 6,000 pounds in its bed? Probably. But whereas the RAM 3500 has the heavier construction and power necessary to tote around 6,000 pounds every day, the 1500 Sport would doubtless experience premature failure. The design simply isn’t built to sustain that level of stress. Similar category distinctions exist in the hard drive world as well.

The difference in rated workload between client and nearline is striking. One has to look at this and wonder how any data center manager might justify running desktop drives under enterprise-grade workloads. The obvious question, though, is whether 180TB/year is enough for cloud environment needs.

The above chart and data from Seagate shows the use results from thousands of analyzed drives taken from bulk cloud environments. As you can see, 75% of all drives examined showed less than 500 GB of combined reads and writes daily. That sounds fairly encouraging for the desktop drive case. However, if we consider the 75% percentile and its 473.4GB/day data point, multiply that by 365 days per year, we get roughly 173TB of annual drive activity. This fits within the workload parameters of the NLL drive but clearly not within the desktop’s parameters, which would appear suitable for approximately 30% of the drives examined.

Seagate data goes on to analyze 1200 drives obtained from Google’s data centers, and one would be hard-pressed to find a better example of cloud usage. Analysis revealed that a “normal” load at Google was 182GB/day (66.4TB/year). One might think that this load is close enough to client-drive specs for desktop drives to serve well in Google’s environment. In fact, 55TB/year means just over 150GB/day. Even this is misleading because cloud environments operate 24x7 and client drives are built for 8x5. A client drive subjected to 55TB/year 24x7 is more likely to fail prematurely than the same drive handling the same workload under 8x5 use. 

Temperature

Temperature compounds workload concerns. Anyone who has ever done manual labor in the sun understands the inherent relationship between ambient environmental conditions and a worker’s ability to perform at optimal levels — or even hit outright failure. Like many other computing components, hard drives have a temperature range in which they operate within normal parameters. As temperature rises beyond this range, failure rates rise. 

Most hard drives are built to run smoothly at temperatures between 5°C and 60°C. These numbers are fairly common on hard drive spec sheets, desktop and enterprise alike, but they don’t tell the whole story. It’s not just a matter of if a drive will operate at the upper end of its operating temperature range, it’s how long it can operate reliably at those levels. This depends on the component tolerances within the drive. Simply put, a desktop drive forced to run continuously at 60°C will fail sooner than an enterprise or NLL equivalent. Clearly, temperature needs to be a primary consideration in data center settings.

Vibration 

In any system environment in which multiple drives and fans are in play, vibration must also be considered. Vibration comes from many sources. Within a hard drive, the spinning motor and arm movements create vibration, even in low-RPM models. Outside the drive, other hard drives, optical drives, case fans, cooling motors, and any other moving components also contribute to vibration within the system. (This is particularly noticeable in storage enclosures and storage servers.) Beyond system components, vibration contributors can include air conditioning, drive hot-plugging, passing traffic, additional nearby systems, and other factors.

Vibration creates pressure waves, and as anyone who has taken high school physics knows, multiple, overlapping wave sources can create cumulative crests and troughs, so even if the average amount of vibration present in a system is tolerable, there can be spikes in which a drive can be subject to occasional, extreme movement. Vibration is measured in radians/sec2, and while all hard drives can reliably write in the face of some vibration, the percentage of successful writes will decrease past certain thresholds depending on the drive’s characteristics.