How Seagate Tests Its Hard Drives

Run Testing

During the Design phase, engineers are challenged to optimize functionality, schedule, and cost without compromising on reliability. Seagate iterates design points until there is confidence that customers will have, as Seagate puts it, “a positive field reliability experience,” meaning that the drive holds up as expected. Does that mean perfect reliability? No. Would anyone have called the Wright Flyer reliable? Of course not. That thing was totaled by a wind gust while parked on the ground following its first day of testing. Yet ongoing waves of design refinement increased reliability, with the third-generation Flyer staying aloft for up to 39 minutes and proving far less prone to crashing.

Seagate’s Design phase flies a similar route, although, unlike the Wright brothers, Seagate engineers know exactly what quality levels they need to reach. Enterprise Capacity, for example, needed to achieve a mean-time between failure (MTBF) of 2,000,000 hours as part of its design objectives. A 2.0 million-hour MTBF is common and expected among a certain class of enterprise hard drives, and the design would not be accepted by the market if it failed to reach this level of reliability.

Two points should be made here. First, engineers could design a drive with much higher reliability. As one of them told us, it’s possible to design a drive that would run for 100 years—but no one would ever buy it. So when you see MTBF specs of 800,000 hours or 2 million hours, understand that those are reliability levels commonly expected by certain market segments. A home PC that might run for three hours per day over five years (about 5500 hours) has no need for a 2 million-hour MTBF drive and thus no need to add the expense of the more expensive parts and additional engineering necessary to meet that metric. A relatively low MTBF does not imply that a drive is inferior. It is merely price-performance optimized for a given market segment.

Second, drives do not begin the Design stage at anything remotely close to their final MTBFs. Just as the Wright brothers were happy simply to fly ten feet over the ground on their first flights, Seagate engineers might only want their first wave of drives to reach a 300-hour MTBF, or even less. Knowing this, they will begin with only a relative handful of drive samples, perhaps 100 or 200, and run them through a battery of basic tests to examine both their performance characteristics and the nature of any failure points. Once that analysis completes, the design will undergo whatever tweaks engineers deem necessary, create a new batch of drives, and try again. Once that first set of reliability and performance metrics is satisfied, Seagate creates a larger population of drives and raises their metric targets, and the cycle repeats.

By the time Design wraps up, a drive will undergo well over 500 tests, many of which take weeks to run.

“Ultimately, we're trying to determine how long will the product live out in the real world when it’s used in normal usage,” explained one engineer on our journey. “The second half to that is design maturity testing, where we push the envelope. We try to see how hot will they go, how cold, wet, dry, voltage margining, scripts. You throw all these parameters at the hard drive in various combinations, and you see where the limits are.”

MORE: Best SSDs For The Money
MORE: How We Test HDDs And SSDs
MORE: All Storage Content

  • tom10167
    Awesome photos. I don't know what the last picture is but I know I need one of those in my house.
  • Rookie_MIB
    Awesome photos. I don't know what the last picture is but I know I need one of those in my house.

    That is an enterprise storage rack full of 2u hotswap chassis. 18 chassis, 12 drives per chassis = 216 drives @ 6tb (?) per drive = 1,296 terabytes or 1.3 Petabytes.

    You could store a lot of TV shows or movies on that thing. Imagine how many of those are used for YouTube? Yikes. They get 300 hours of footage uploaded every minute.
  • Steelbridge
  • Mike-TH
    So if their testing is so good, why are their drives among the worst for reliability - to the point where most IT people I know actually refuse to use them, or if forced to use them will keep (and use) more spares than for other makers.
  • Tom20160027
    The article explains the different types of drive/MTBF and why the backblaze test is useless information. Marketing plot to have folks talking about it and re-posting its link. It seems to work as we keep seeing the link over and over... They are not getting my data. They put drives designed for desktop into servers and run them to the ground and call it a "reliability test". Let's test my kids bicycle with training wheels at the Tour de France and complain about its quality....

    I know IT folks that refuse to use other brands of drives as well. I know IT folks that refuse to use servers from this brand or that brand. We can find anecdotal information about anything. It does not make it true.
  • Glock24
    Seagate tests their drives? I thought they didn't!

    I've had more Seagate drives die without warning than any other brand. The only ones that have survived are some old 250GB Barracuda ES. All other models I've owned had lots of bad sectors or just stopped working before the first year, but SMART almost always says the drive is fine!
  • zodiacfml
    Yawn. All I think of right now is that HDDs will become the tape drives of the past.
  • Garrek99
    The only drives I've ever had go bad on me were Seagate drives.
    Every other drive I've ever purchased simply became obsolete due to size and thus replaced.
    They should be reading about how the other drive makers do their testing and learn from that. Hahaha
  • rosen380
    Maybe things changed... but all of my old SGI machines always had Seagate drives in them and the 20+ year old drives all still work. Hell look at what these drives *sell* for on eBay:

    4.5 GB drives *selling* for $150+ I see a 2Gb for $120.

    They must have been pretty decent at some point if SGI was putting them in their $5000-20000 workstations and people are spending $40+ per GB to get these now...

  • rosen380
    Link was too long...