16
votes

Wiki for information on which storage manufacturer(s)'s drives have the longest life / lowest failure rate?

Everyone has a favorite, but if you can refer to a (hopefully recent and somewhat impartial) study, that would help avoid subjectivity.

EDIT: Any information is useful, so rather than limit the question to server-grade drives, or a particular size etc I'd just ask that you mention any particulars in the answer.

4
  • 10
    If you haven't already read Google's excellent study on generic drive failure rates (which most of you probably have) see labs.google.com/papers/disk_failures.pdf
    – username
    May 13, 2009 at 20:16
  • 2
    Laptop disks, cheapo SATA disks, high-performance desktop disks, enterprise SCSI/SAS disks, SSDs, some form of smeared-across-all-types-pointless-average? which?
    – Chopper3
    May 13, 2009 at 20:27
  • Any good info you're aware of could come in handy to me (and others). I'm mainly asking to get a sense of what's out there. My coworkers obviously have favorite brands - unfortunately opinions on this type of thing are often not based on facts, but on hear-say, or an experience someone had 10 years ago, etc
    – username
    May 13, 2009 at 20:52
  • google uses (or used) hitachi deskstar in their servers Dec 13, 2012 at 9:05

10 Answers 10

18
votes

Interestingly, Google published a study that looked at a pretty large population of drives and their failure rates. One section says, and I quote:

"In contrast to age-related results, we note that all results shown in the rest of the paper are not affected significantly by the population mix."

The authors found that drives tended to fail (1) very soon after their first reported SMART scan error and (2) based more on the drive model number and vintage rather than manufacturer.

1
  • 2
    I just read part of the paper, and it states "Failure rates are known to be highly correlated with drive models, manufacturers and vintages. Our results do not contradict this fact." Manufacturer is in the mix - it seems to be at least somewhat important.
    – Phil
    Aug 12, 2011 at 15:01
7
votes

The issue with this question is that there are good drive models for a manufacturer and bad drive models.

For example, there seems to be a high failure rate for the Seagate 1tb & 1.5tb drives, but their 250gb & 320gb drives are solid*. Yet the Western Digital 1tb, 1.5tb, & 2tb drives seem more stable, but there have been issues with their 500gb drives*.

The answers to this question should address more of the good drive models, not the actual manufacturers themselves.

Thanks! JFV

**Information for this data was obtained from ratings on websites like NewEgg.com, TigerDirect.com, etc*

4
  • 1
    bugger, just ordered a 5h1tl04d of those 1.5TB disks :( thanks anyway
    – Chopper3
    May 13, 2009 at 23:52
  • @Chopper3: I almost did too (because of a great price), but did some research and found that they (the Seagate 1.5tb drives) have been failing quite a bit!
    – JFV
    May 13, 2009 at 23:55
  • 1
    IIRC, the big Seagate drives also have been having some issues with their firmware.
    – Tim
    May 14, 2009 at 1:45
  • 500GB Seagate drives from early 2009: 7200.11, DiamondMax 22 and Barracuda ES.2 SATA hard drives: mswhs.com/2009/01/21/seagate-hard-drive-firmware-bug
    – username
    May 14, 2009 at 6:53
4
votes

StorageReview.com stores reliability rates for hundreds of drives. You can view by manufacturer and see which drives have been reported to be reliable or failure-prone. The data is mostly user submitted (perhaps even by SF peers) with 52900 entries to date. At least you get numbers:

StorageReview's Survey

Registration is free, and you must enter at least one drive result but you get to view the entire database.

EDIT: Regarding Google, they used Hitachi Deskstars in one of their Data Centers at one point: alt text http://www.hyperslug.com/image/photo/GoogleServerLarge.jpg Courtesy of CNET

3
votes

I hear them Winchester drives are pretty good.

2
votes

Before you can understand the answer, you must first understand the question (ooohh, I durn sound smart!)

Seriously, you have to define a "failure" before you can really ask the question.

Is a failure:

  • Failing to read the disk and returning bad data?
  • Failing to read the disk and returning a read failure code once?
  • Failing to read the disk and returning a read failure code X times?
  • Total loss of access to your data? (i.e. head crashing, electronics failing, spinup fail)?

Of course, in addition to the Google report that username has linked, NetApp's whitepaper on data corruption is IMO a must read for any storage or systems administrators.

2
  • Based on your first sentence, I was expecting the rest of your answer to just be "42" May 13, 2009 at 23:19
  • I'd be happy enough with reliable studies for any of those modes of failure - it's not like we're going to conduct the study, so we can let the study authors define failure for us. May 14, 2009 at 0:27
2
votes

There aren't that many disk manufacturers these days, I'm sure we could all name 90% of them right away - I honestly don't think there is a true answer to this question - not across the board and not one that we can all agree on.

Ultimately this comes down to the fact that no one company is generally perceived as being of significantly lower quality than the others. I'm sure there have been single-model/batch issues with all of the main players but overall they're all pretty competitive with each other on speed, capacity, reliably etc.

So the bad news is that I'm not sure there's an answer to your question but the good news is it's this lack of quality tiers means that these days we can all get very fast and reliable disks for comparatively very little - which is nice :)

2
  • If you can name 90% of them then that implies there are at least ten manufacturers :-)
    – John Fouhy
    May 14, 2009 at 1:54
  • Arf! good one, got me there :)
    – Chopper3
    May 14, 2009 at 7:49
1
vote

A more serious answer from me is, perhaps you're asking the wrong question :)

Google figured out a long time ago that you must design for failure. For the single user, I would recommend something like a Drobo combined with a good backup regimen. Server-side, it is not that much different, except that rather than a Drobo you probably want to look at RAID solutions. For applications, you can also consider something like cloud-storage, e.g., Amazon S3.

2
  • right... so what brand of server-grade hdd's should i use in my raid? i don't like ordering spares every two seconds ;-)
    – username
    May 14, 2009 at 5:38
  • 1
    I think the answer is, the cheapest and have lots of redundancy.
    – RedFilter
    May 14, 2009 at 15:34
1
vote

I have a nice graph (under NDA, sorry) that shows that Enterprise class SATA drives from Seagate and Hitachi are all slightly below the 1% annual return rate for the first 3 years, in case that helps.

You could also check the Storage Review reliability survey and see what they say of your favorite vendor or drive. However I don't know if this is of any statistical significance.

0
votes

Since Google (in the study mentioned above) is the only one to have performed a field analysis with a sufficiently large population size, (more than one hundred thousand disk drives) it would be interesting to know which drives they now purchase.

Any IT flies on Google's walls out there have some comments?

Unfortunately for the rest of us, they were unable to name names of manufacturers (or specific models) with the results. Instead (frustratingly) they confirm what we all already know, namely that "Failure rates are known to be highly correlated with drive models, manufacturers and vintages."

I tend to agree with JFV above - there are good drive models for manufacturers and bad drive models for the same manufacturer. Things change over the years too. It is therefore better to focus on models rather than manufacturers or brands.

Pay attention to warranty offered and anecdotal feedback on specific models from the likes of newegg, tigerdirect, amazon.

As the Google study seems to confirm, posted MTBF is often less informative than many people assume.

0
votes

I have used thousands of drives over the years and about half tend to fail within the first year. This is especially true for servers or desktops that have large amounts of disc activity. Seagate, maxtor and western digital are the brands I use most of the time but I have learnt that buying large numbers of identical drives at the same time is suicide.

My solution is to spread the risk by purchasing between 2 and 5 at a time - constantly changing the manufacturer, the size and the supplier. This spreads the failures out over time and it makes it easier to figure out which manufacturer has messed up.

Capacities increase as technologies change and although the manufacturers will do a lot of lab testing, this is not the same as regular drives in regular use.

The fastest way to kill a large hard drive for consumers is having a small amount of physical RAM in the PC. When windows XP came out 256mb was generous but now 1gb is minimum and to prevent hard drives dying to quickly you need 2gb for heavy users. When switching to 64 bit operating systems you need 2gb minimum and 4gb to be comfortable but heavy users will need 8gb of RAM. Running a wordprocessor, an email client and a browser at the same time - and this is not an unusual - will be swapping memory like crazy on a machine with small amounts of ram and the sweet spot over the years keeps going up. It it not unusual to require 1gb of RAM for the browser alone now.

2
  • 2
    No. Just... no. Dec 12, 2012 at 15:14
  • Seriously? Half? That sounds like quite an exaggeration.
    – cmorse
    Sep 11, 2013 at 23:10

Not the answer you're looking for? Browse other questions tagged .