Hard Drive Reliability Stats for Q1 2015

By | May 21st, 2015

Hard Drive Reliability Statistics

Over the past year we’ve been releasing hard drive reliability statistics based on the drives we use to store customer data for our online backup business. As of the end of Q1 2015 we had 44,252 hard drives spinning in our datacenter. If we subtract boot drives and drive models with less than 45 drives from that total, we get 42,749 hard drives remaining spread across 21 drive models. Below are the hard drive reliability statistics for these drives for Q1 2015.

Backblaze Hard Drive Failure Rates January 1 – March 31, 2015
Name/Model     Size Number
of Drives
Average Age
in years
Failure Rate
95% Confidence
HGST Deskstar 7K2000
2TB 4,664 4.1 1.66% 1.0% – 2.6%
HGST Deskstar 5K3000
3TB 4,596 2.9 0.79% 0.4% – 1.5%
HGST Deskstar 7K3000
3TB 1,013 3.3 2.00% 0.6% – 4.7%
HGST Deskstar 5K4000
4TB 2,653 2.0 1.07% 0.4% – 2.2%
HGST Megascale 4000
(HGST HMS5C4040ALE640)
4TB 7,026 0.7 0.80% 0.4% – 1.4%
HGST Megascale 4000.B
(HGST HMS5C4040BLE640)
4TB 3,100 0.9 0.39% 0.2% – 1.10%
Seagate Barracuda 7200.11
1.5TB 259 5.0 31.68% 20.1% – 47.5%
Seagate Barracuda LP
1.5TB 1,485 5.1 12.16% 8.9% – 16.3%
Seagate Barracuda LP
2TB 81 2.8 0.0% 0.0% – 33.9%
Seagate Barracuda 7200.14
3TB 485 2.4 26.65% 19.6% – 35.4%
Seagate Barracuda XT
3TB 234 3.2 3.05% 0.4% – 11.0%
Seagate Barracuda XT
4TB 175 1.9 4.62% 0.6% – 16.7%
Seagate Desktop HDD.15
4TB 14,803 1.0 3.06% 2.5% – 3.7%
Seagate 6 TB SATA 3.5
6TB 495 0.1 2.41% 0.1% – 13.5%
Toshiba DT01ACA Series
3TB 50 2.0 5.1% 1.6% – 11.8%
Toshiba MD04ABA-V Series
4TB 45 0.1 0.0% 0.0% – 106.9
Toshiba MD04ABA-V Series
5TB 45 0.1 0.0% 0.0% – 106.9%
Western Digital Red 3 TB
3TB 1045 0.9 12.87% 8.5% – 18.7%
Western Digital 4 TB
4TB 45 1.0 9.01% 0.2% – 50.2%
Western Digital Red 6 TB
6TB 450 0.3 7.77% 3.4% – 15.3%
Note:  Some of the HGST drives listed were manufactured under their previous brand, Hitachi. We’ve been asked to use the HGST name and we have honored that request.

Hard Drives In and Hard Drives Out

In our most recent hard drive study for Q4 2014 we analyzed 39,696 drives. There are 3,053 more drives in our Q1 2015 results. We actually deployed over 4,000 drives, as we replaced 324 failed drives as well as 631 Seagate 3TB suspect drives. During Q1 we deployed two Backblaze Vaults containing 900 drives each and 44 standalone Storage Pods with 45 drives each.

Observations and Trends

Looking at the Q1 hard drive reliability stats above you can compare failure rates for the various drive models. The important thing to remember is that the table is a snapshot in time: Q1 of 2015. Below are the cumulative failure rates for the same 21 drive models beginning with Q4 of 2013 and progressing quarterly up through Q1 2015.

Backblaze Hard Drive Failure Rates, Cumulative by Quarter
Name/Model Size 12/31/13 03/31/14 06/30/14 09/30/14 12/31/14 03/31/15
HGST Deskstar 7K2000
2TB 1.10% 1.08% 1.09% 1.03% 1.06% 1.15%
HGST Deskstar 5K3000
3TB 0.90% 0.85% 0.70% 0.73% 0.74% 0.74%
HGST Deskstar 7K3000
3TB 0.90% 1.54% 1.46% 1.55% 1.81% 1.83%
HGST Deskstar 5K4000
4TB 1.50% 1.33% 1.25% 1.06% 1.17% 1.16%
HGST Megascale 4000
(HGST HMS5C4040ALE640)
4TB 2.67% 1.90% 1.86% 1.43% 1.18%
HGST Megascale 4000.B
(HGST HMS5C4040BLE640)
4TB 20.29% 1.23% 0.59% 0.52% 0.48%
Seagate Barracuda 7200.11
1.5TB 25.40% 22.27% 22.98% 23.02% 23.41% 24.12%
Seagate Barracuda LP
1.5TB 9.90% 9.87% 9.67% 9.56% 9.93% 10.18%
Seagate Barracuda LP
2TB 7.20% 8.03% 8.18% 9.96% 9.63% 9.93%
Seagate Barracuda 7200.14
3TB 9.80% 13.92% 17.65% 27.15% 28.31% 28.26%
Seagate Barracuda XT
3TB 7.30% 6.53% 6.33% 6.08% 5.59% 5.27%
Seagate Barracuda XT
4TB 0.75% 0.56% 0.45% 1.12% 1.61%
Seagate Desktop HDD.15
4TB 3.83% 3.03% 2.73% 2.75% 2.83%
Seagate 6 TB SATA 3.5
6TB 1.70%
Toshiba DT01ACA Series
3TB 4.63% 3.48% 4.20% 4.81% 4.23%
Toshiba MD04ABA-V Series
4TB 0.00%
Toshiba MD04ABA-V Series
5TB 0.00%
Western Digital Red 3 TB
3TB 3.20% 8.78% 9.07% 6.96% 6.49% 7.90%
Western Digital 4 TB
4TB 9.01%
Western Digital Red 6 TB
6TB 13.75% 3.07% 6.64%

The cumulative view can be used to see trends over time as well as spot anomalous periods requiring further review. Remember that in the first quarter a given drive is deployed the results can be skewed due to the limited number of drives and deployment hours.

Combining Views

Internally we combine the cumulative and the current failure rates for each drive into a graph so we can quickly see if there are any changes requiring our attention. Below is a sample showing this information for the two 6TB drive models we currently deploy. The “overall rate” is the cumulative failure rate since the drive model was put into service and the “recent rate” is the failure rate over the past 2 months.

Drive Stats Tracker

The data for Q1 2015 is available from our Hard Drive Data web page where it can be downloaded for further study. The format for the 2015 data file changed with the addition SMART 22 attribute (Raw and Normalized) and 4 additional SMART stat pairs. Adjust your schemas accordingly and let us know if you find anything interesting.

Andy Klein

Andy Klein

Director of Product Marketing at Backblaze
Andy has 20+ years experience in technology marketing. He has shared his expertise in computer security and data backup at the Federal Trade Commission, Rootstech, RSA and over 100 other events. His current passion is to get everyone to back up their data before it's too late.
Category:  Cloud Storage
  • In Service 5-5-14
    4- 4TB Deskstar Coolspin 3.5″ SATA III Internal

    2x Seagate Desktop HDD.15 ST4000DM000 4TB 64MB

    2x Seagate Expansion 3 TB USB 3.0 Desktop External Hard Drive STBV3000100

    Western Digital Caviar Green 3 TB 5400 RPM SATA III 64 MB Cache Bare/OEM Desktop Hard

    We had been buying WD drives for years. But after flooding ran drive prices up, Seagate was the first to have prices become a bit more reasonable, so as you can see we bought four Seagate drives.

    Withing two weeks of each other the WD 3 TB failed, and one of the Seagate drives died outright. Testing of the other three Seagate drives showed abnormal error rates.

    So I bit the bullet and bought Hitachi drives and junked the Seagate drives, at a cost of nearly a grand and hundreds of n=man hours making the switch.

    That kind of thing only happens once with a vendor, and they are toast. So while Seagate was the primary cause to switch vendors, I tossed WD because their drives are less reliable than Hitachi and because the WD drive should not have failed that soon.

    The issue is reliability and not losing data. The combination of the two drives failing so close to each other cost me data, and about $30,000 to recreate just part of the lost data.

    From then on Seagate would never sell us another drive. They make unreliable junk.

  • iTwns

    Funny, I always think so highly of Seagate and very negatively on WD. Now that I read your reports, my legs are shaking as I use my Baraccudas to store my most valuable data. Now that I think about it, 2 of 4 of my ST2000M001s have died and at least one WD 2TB died (died yesterday). None of the IBM/Hitachis have ever died on me. Three of the 8+ years old 1TB IBM Deskstar drives are still running 18 hours a day. I am now writing this on the subway heading to the computer store to buy HGST drives to replace the Baraccudas and WDs. Thanks for the extremely useful reports.

  • jk

    Amazing – thanks for you hard work.

  • Robert James Crawley Klopp

    After my friends Seagate died and caught on fire at the same time I am never buying one again, especially when it almost killed my beloved desktop.

    I have a 2TB WD Green sitting in my cupboard at the moment and it still works to this very day. And i’m also surprised that a 18yr old HDD is still working and it’s a Maxtor the one that is a failure.

    Also ripped from another computer was a Seagate was a 1 and a half year HDD, but it was also too dead. My most trusted brands are WD, Smasung and Hitachi. Wouldn’t mind Hitachi or Fujitsu, but they’re getting way past my curfew, so now it’s time to pull out my big guns of speed.

    I’ve also noticed none of my drives developed faults in them during the course of certain doom with my MOBO and they’re still here by my side working.

    I wouldn’t think it’s reliability that causes this. I would possibly think about how much data is on the persons Hard Drive and what they’re using it for (big files can cause the hard drive to wear down over it’s lifetime, therefore making the drive fail would be that it’s being stressed pass it’s limit), but then again Seagate is on the bottom of my list.

    But guess what the Xbox One is certain doom for? Yep it’s got a Seagate HDD in it. And is really a bad brand to pick up. After my incident I am lucky to survive and to have 1000’s of dollars of my stuff still here.

    This gives me a good gap to sue Seagate for a Safety Hazard with the Hard Drives they make.

    My theory might not be correct, but at least some of you understand what i’m saying correct? I wonder what i’m going to do with all the GB’s I have at the moment? I have got at least 4tb of empty space waiting to be filled up with who knows :)

    I am going to tell you a very short story:

    I had a 32GB USB in tip top condition, but one day it wouldn’t be recognised by Windows. I thought the drive was stuffed and it looked like it. Until I plugged it in to my Linux PC and BAM recognised it straight away.

    I would not go out of the 2TB barrier I set upon my self as higher capacities will have higher failure rates. It would seem like i’ve arrived in a modified world of No Man’s Land. If I had the 1TB I had today from a year ago I would’ve been very happy. Alas I dropped it and the click of death overwhelmed me. Simply put I would’ve cracked the enclosure and fixed it myself, but if it happened to the new one I have which I would never do, then come out the gloves and in a clean room I go.

    Anyway too much storage is a no for me, except my 2TB HDD. which holds my Naruto and One Piece anime.

  • Tan Lek Tiong Alvin

    Hi can anyone recommend a HD to get for my QNAP TS-451? I compared what is recommended through QNAP and compared it with


    Some people are not recommending Seagate as they have high failure rates and come to some conclusion on 2 brands, HGST Deskstar NAS/Enterprise Drives H3IKNAS40003272SN or Toshiba Surveillance Drives MD04ABA500V.. it seems backblaze has no data on the models QNAP listed and only has the hitachi drives. it has good performance and 0 failure rate so far. But the data is only based on 45 HDD drives compare to HGST. It seems HGST is an industry standard?

    What I cant find is anyone using surveillance drives to stream movies and TV shows… I usually leave my Home sever on and watch movies and shows in my room. Would it be ok if I use it for media storage and streaming instead of video recording?

  • Andrew

    Thanks for the great info. i have two Hitachi Deskstar HDD’s in my HP Box. The date on them is Dec 2007. Never had an issue. All I can say is I look for quality and Hitachi is it for me. Thx.

  • Andy Diamond

    I don’t understand how you compute the confidence intervals, Example:

    Seagate Barracuda LP
    (ST31500542AS) 2TB 81 2.8 0.0% 0.0% – 33.9%

    I see the sample size is small but 81 is not all that small neither is 2.8 years so how do you get a 33.9% upper limit on your confidence interval? Or should I write, if that’s not a confidence interval on annual failure rate what is it and how are you computing whatever it is anyway

  • str

    What exactly constitutes a “failure”? There doesn’t appear to be a definition for the purposes of this data set. Were failures determined solely by SMART status? Or by other means?

  • N GeForce

    What does a 95% Confidence Interval of x% mean?

  • Zephroth Derusus Talinguard

    Can Confirm the Seagate HDD failure rate, this includes the 500GB and 1TB models, unsure about the rest. Out of 10 systems we have had 5 Failures. And will prolly have more by the end of year.

  • cyrylthewolf

    Seagate is PURE GARBAGE. They’ve been a nightmare at times in the data center. (We’re one of the big three. It’s one of the largest data centers in the world.)

    I wonder if anyone else here has experienced the cascade failures of the “MOOS” drives… They continue to plague us every now and then. (Although I think that we’ve got most of them out of our filers.)

    And then, in my own home tonight, I have BOTH of the Seagate drives in my RAID1 fail. Turns out that the 500GB OS drive (also Seagate) is suddenly on it’s last leg as well.

    Really should have bought some HGST’s sooner than this. Now I’m doing data recovery from backup in the wee hours. :P

    I will never go near another Seagate drive so long as they exist.

    • johnkristian

      I had 15 Seagate ST3000DM001 in one of my fileservers. They died ALL THE TIME. It ultimately cost me 4800USD to replace them with WD RE4 drives.

      • cyrylthewolf

        That’s funny. That model number is interestingly close to having “M00S” in it. Except it says “M001.”

        I’m not surprised at all. Sorry it cost you so much, man. I feel your pain.

        • Stoatwblr

          I’m seeing similar issues with ST2000DM001 drives. They keep popping off the bus for no apparent reason (and ALL of them have been replaced under warranty between 1 and 2 years)

          It’s interesting that BB are seeing low failure rates on the ST4000DM001s. but I have to wonder if this is because they’re early in their life cycle.

        • Richard H.

          That’s funny? No, it’s not. Am I banned from this shithole yet? You fucktards are boring.

    • Richard H.

      You’re familiar with Seagate’s alleged failure rate because of your highly credentialed anecdotal data, yet use at least three of them in your own home.

      Yeah, sounds legit lol

      • cyrylthewolf

        Don’t be dense. Of course I don’t. Why would I? I had them before that job. Learned during the course of that job just how terrible they truly are. Then I experienced it for myself. They’ve all been replaced since then – with prejudice – by HGST drives as well as a new RAID controller upgrade.

        Thinking for a moment might have led you to consider that as a basic, logical conclusion…MAYBE.

        Is it really so hard to think BEFORE speaking? Or is it just that it’s so much easier to troll people for no good reason?

        Fool troll.

        • Richard H.

          Capitalize a few more words, CyrilTheAngryPuppy. That’ll SHOW everybody you mean BUSINESS.

          • Alexander Craig

            Yep, you are definitely a troll. Nothing in your comments about the subject, just vitriol. Go play somewhere else, eh?

          • Richard H.

            Speaking of adding nothing to the conversation…thanks for your keen insight :)

          • cyrylthewolf

            Tempted to say simply that the concept of EMPHASIS clearly escapes you. Perhaps that is the case either way.

            But the more logical conclusion is that you’re an immature little child trolling a website where industry professionals gather to learn and trade information.

            Why are you here again? You can’t possibly be one among us.

            Stick to the topic or GET LOST.

          • Richard H.

            The “industry professional” came back to a three-week-old comment to reply to a troll post. I bet you’re a big huge fat guy, because clearly you have no self control.

            Knock yourself out with the last word, fella. I’m sure the REST of the professionals are dying to read your GENIUS wit, complete with RANDOM emphasis. Fucking tool lmao

          • Rockman Rock

            You fail harder than Seagate.

          • Say What??

            You’re an idiot Richard….a classless troll with no manners. Your parents obviously taught you well. People like you need to shut your mouth and stay off the Internet and get a real hobby that involves getting out of your mom’s basement.

  • waltc4

    Since I see no 512GB drives in the list, I’ll assume (you know what they say about that) that my twin Seagate RAID 0 512’s at home are likely probably good for another 3 years, hopefully (as they’ve already lasted four years as it is)…;) I hope…

  • Илья Василевский

    I have extracted just the failures (rows where failure column contains 1) from all the data on this page:


  • Илья Василевский

    Hi :) I downloaded and extracted 2014_data.zip two times to be sure and I got this error both times:

    inflating: 2014/2014-03-17.csv bad CRC 933d0cb7 (should be f5868a9b)

    Consequently, this command fails:

    csvgrep --columns failure --match 1 2014/2014-03-17.csv
    list index out of range

    Looking at the file, I can see that it’s garbled starting from line 23520:


    Could you repackage this file? Or the whole 2014 archive?

    • It might be a poor internet connection? Seems to work on our end? Though, reading your above comment it seems like you’ve got it downloaded and working!

      • Илья Василевский

        Other location, other computer, other connection. Problem remains:

        I was able to process the bad file after deleting the garbled lines, which is written in the script I provided in that gist.

  • Илья Василевский

    Hi :) I’m downloading the archives from https://www.backblaze.com/hard-drive-test-data.html and I noticed that the server doesn’t seem to support partial downloads? `wget -c` fails and downloads the file from the beginning :( Thought this was something you might want to fix…

  • This is a very useful and much needed overview of the hard drives. While purchasing any hard drive one goes through customers’ reviews, brand value of the product, price, etc. but has less information about the performance of the product. The customers’ reviews, though unbiased, are also not able to provide such deep insights about the performance of the HD over prolonged usage. Like for instance I did not know that Seagate and Western Digital hard drives have such high failure rates. Thanks for the info! It will help a layman take better informed decisions as far as HD are concerned.

  • chiniwalar

    sir, nice explaination overall. Thank you.Appreciate your patience. please suggest me any 1 TB internal desktop HDD, which has long lifespan without failure.(> 6 years?). I live in INDIA.

  • VampireDuck

    Thank you for this, it was quite interesting. However I am looking for a 1TB HDD but they havn’t been tested here, so I was wondering if anyone could recommend to me a reliable 1TB HDD?

    • johnny

      My personal user experience has been with an internal drive of 1TB SAMSUNG HD 103 UJ ATA drive
      it lasted from 2009 and still going. But that;s a single user exp. What you need is statistics for that drive. Search on Amazon for reviews and the like.

      A seagate of 3 TB just failed on me one week ago :(
      Trying to get data back and decide on a new HDD…
      From the graph, I see I shouldn’t buy a 3TB again. Either lower storage below 1TB x 2 drives or 1 drive with 4TB.. (which seems to be a more reliable storage than 3 TB,even more than 1.5 TB)

  • Rick Peralta

    Excellent raw data!

    It definitely dispels the hype about reliability.

    I should think that the drive manufacturers would woo Backblaze to use their drives… if only to get actual field data.

    On a side note, with over 40K drives spinning, what does that look like? An ocean of racks? A megawatt power feed. What about power backup? Have you written about all that?


    – Rick

  • Alberto Ochoa

    tl;dr :-) Which one is the most reliable overall?

  • Yev et. el.

    I love you guys. That’s all


    • Hah! Thanks doc :D

      • cyrylthewolf

        Yeah. You guys ARE pretty great. I can’t find a more definitive source of good, solid, researched info anywhere. (At least…nothing that I like or isn’t biased.)

        I’ve tried to show some of the folks your graphs in the company to back up what we’ve experienced in our data center. Can’t wait for the next one.

        Thanks for the hard work, guys. :D

        • Richard H.

          Nobody CARES what you like. Or what you THINK. Fucking moron lmao

          • cyrylthewolf

            @YevP:disqus – Pardon me. Is there any way that we could boot Richard H. from the thread? He’s clearly not here to yield any relative, productive discussion.

  • Jmaharry

    Amazingly useful info. Thanks! The Deskstar performance is really impressive. Is there a simple way to determine which hard drive brands use these exact high performing deskstar drives in their enclosures?

  • Dan

    Well , the review is good.
    It gives general terms.
    To my taste it is a bit too vague.
    All the big vendors (Seagate , WD , HGST) have 3 relevant categories of drives:
    1.Desktop and Mobile (usually in 2.5″ form , lowest MTBF , cheapest price).

    2.NAS (3.5″ usually but not always) – WD RED and RED PRO , DESKSTAR NAS for HGST and SeaGate NAS +Enterprise NAS .

    3.Enterprise (WD – Re Series , HGST 7KX000 series , Seagate Enterprise capacity).

    I would prefer a review of all inside their own sections.
    Currently it is not an apples to apples comparison which makes decisions very hard and data very vague.

    Do you have any plan of doing a few sections according to the divisions as offered above ?
    It would be very helpful for anyone reading the article.


    • Wow… How about just thanking them for helpfully providing all of their data periodically, and if you’d like to see it broken down further, maybe you could do it, and share your results with them? They give all of the model numbers.

      Thank you Backblaze for tracking and releasing all of this information. I refer people to your HD data constantly.

      • Corey Kronengold

        wow…maybe you’d just read that as as suggestion for more specific breakdown of the data. Personally, I was looking for the exact same things.

        While this info is certainly handy, it would be a very welcome addition to have this broken down between desktop, NAS and enterprise.

    • Edward Wyrwas

      As someone who performs failure analysis on various hard drives and solid state drives, I agree about the classification of the disk drive statement. The utilization duty cycle, expected performance, and number of powered on hours, etc. all need to be taken into consideration. Also, on the topic of reliability, it would be nice to see how BB’s vertical installation of these drives affects the vibration response of the disk drive itself and also the air flow/case temperature of the drive. It’s never apples-to-apples unless you take these things into consideration.

      A clip from your website on vibration: The drives vibrate too much if you leave them sitting as shown in the picture above, so we add an “anti-vibration sleeve” (essentially a rubber band) around the hard drive in between the red metal grid and the drives. This seats the drives tightly in the rubber. We also lay a large (16″ x 17″ x 1/8″) piece of foam along top of the hard drives after all 45 are in the case. The lid then screws down on top of the foam to hold the drives securely. In the future, we will dedicate an entire blog post to vibration.

      So the foam holds them in place from the top, adding backside support to keep it connected to the backplane… and essentially a rubber band to reduce some vibration to your framework which holds them vertically…

      Don’t get me wrong, I am a fan of what Blackblaze is trying to do in regards to bringing affordable storage to the masses. But for disk drive reliability, I have my concerns. Most notably, I am very glad you’re not using SSDs.

      • Coenraad Loubser

        You can get that off the SMART data that they publish on their Hard Drive Data page! Go… analyze it for us!

    • Edward Wyrwas

      More information: Desktop HDDs tend to have a duty cycle of 8 Power On Hours (POH) per
      day. Enterprise drives tend to have a duty cycle of 24 POH/day. A lot
      of NAS drives are spec’d at 1 POH/day (8760 POH/yr).

      Apply these duty cycles to the failure rate of each drive population and adjust for how many actual years you’ve consumed from the drive health.

  • WSP

    Was there adjustment for the amount of actual use? Just because two drives are in your system, does not mean that the actual quantity of reads & writes, number of sectors used, etc. are the same for both of them.

  • Tik B Lang

    So, is it still the HGST 4 TeraByte the most reliable SATA drive in the planet? I’m thinking of buying drives and [DISLIKE] hate the ST3000DM001 drives which is the most unreliable drives in the market.

    • johnkristian

      I had 15 Seagate ST3000DM001 in one of my fileserver. They died ALL THE TIME. It ultimately cost me 4800USD to replace them with WD RE4 drives.

      None of my HGST 7k4000 have failed. :)

  • FollowTheORI

    Thanks for the effort put into this and the regular reports on data/stats. Good work and keep them coming! :)

  • Kyle

    I don’t get how that graph shows the Seagate 6TB drive has a higher “Recent” failure rate than the WD 6TB model. The table above the graph shows the Seagate’s most recent quarterly failure rate was 1.7% and the WD’s most recent failure rate was 6.64%. In fact, in every quarter in that table the WD 6TB has a higher failure rate than the Seagate 6TB. Can someone please explain?

    • Andy Klein

      The annual failure rate chart was taken on May19th while the “table” data was as of March 31st. During the intervening period there were several Seagate drive failures which were reflected in the chart. We put the chart in the post to show how we can sometimes see trends before they become problems, but we didn’t give the proper date context. Sorry for the confusion.

  • Casparvl

    Very useful overview! Though techsites often have nice reviews about tech-products, one of the things that is notoriously hard to test is hard drive reliability, as it takes many disks and many months/years of testing to draw any meaningful conclusions. This is where your results are extremely helpful! In my particular case, I was looking for which drives to use in my home-NAS :)

    I myself am a tech-enthousiast, but I work in the medical field. Here, we do many survival analyses, which analyze the time until a certain event occurs (e.g. relapse of a patient, or, in your case, drive failure). A very common way for us to display how long patients remain symptom-free after a certain time is by creating Kaplan-Meier curves http://en.wikipedia.org/wiki/Kaplan%E2%80%93Meier_estimator which can account for different follow-up times (e.g. your drives may not all be installed at the same time, and therefore each have a different follow-up time). Using log-rank tests http://en.wikipedia.org/wiki/Log-rank_test one can test whether survival curves are significantly different. Additionally, one can use Cox-Regression to test whether different variables (such as temperature, like you guys tested for this blog https://www.backblaze.com/blog/hard-drive-temperature-does-it-matter/ ) significantly influence survival. Any book on medical statistics will likely include explanations of these methods and such analyses may be readily performed with software like SPSS or R (the latter has the advantage of being non-commercial free software :)).

    You might already be familiar with the above methods and might already be using them, but just in case you’re not and might be interested, I thought I’d mention them :)

  • Stan Brinkerhoff

    It would be nice to see the pod revision included with each drive failure. Backblaze has noted that the vibration mitigation work specifically on the Pod 3.0 has reduced failure rates. Is there a correlation between Pod revision and failure spread? Do Seagate drives fail more in Pod 1.0/2.0?

    • We have only have some anecdata. But if you look at the HGST drives, we’ve been buying them throughout all the numbers and their failure rates are pretty consistent. That’s likely the case throughout the other drive models as well, but we don’t have the data on that.

      • Sedokun

        I’m mainly WD user, but I’ve got interested in Hitachi’s disks.
        Do You have Data on 5-6Tb models?

        Also, I’ve read a Temperature/Reliability article, and was wondered, all the data is below 30C. But many disks heats up to 40-50C on air (no case/fan, just lying on table), I know that most datacenters are like refrigerators, but what about SOHO usage?

        • johnkristian

          I run about 30 WD RE4 (15x 4TB, 16x 1TB) + at my home NAS + SAN. I also started to buy 4TB HGST 7k4000 drives for a fileserver when they came out, and I’m very impressed. I’m planning to extend this volume to a larger array or multiple arrays in a few months, and I’m definitively will continue buying enterprise HGST drives.

        • Edward Wyrwas

          They should do case temperatures of the drives within the array. Reliability decreases by a factor of two (2) for every 15 degrees the drive is above 25C-30C.

  • Tristan

    Thanks for the data.
    Where’s the graph with all the data? I don’t see it

    • Tristan, we didn’t make one this time around, just released the raw stats. Seems like that was popular though, we’ll consider doing it again if it’s “in demand” :)

      • Tristan

        Oh, that’s fine. Makes it easier to compare the data, though.

        Thanks for your work, that failure % on the barracudas reflects my personal experience with them

        • Yea it’s not a great performer, but the 4TB are working out better!

  • gotthumbs

    Thanks Andy, Some of us do find this information helpful/informative. Especially when we read all the details/information you’ve provided. I do find this information very interesting reading and helpful for when I make my next HD upgrade to my NAS. Thanks again and keep up the good work.

  • Pillion Mac

    Hitachi / HGST are same Manufacturers ? the Model / PN are same ?

    • frogstein

      HGST is Western Digital.

      • Ryan

        HGST is NOT Western Digital. They are just owned by them now. They still do their own products based on their previous successful designs. I would buy HGST branded product over WD any day. I used to work there for a few years. We always operated as our own company within the WD family.

        • frogstein

          I stand corrected. Apparently the purchase agreement stipulated that they remain separate entities. Whether that can be enforced long term remains to be seen.

        • Stoatwblr

          HGST is owned by WD, however the chinese anti-monopoly regulator won’t allow WD to fold HGST within the organisation as it would result in an effective market duopoly (toshiba don’t ship enough drives to matter)

          This is a good thing. The day they are folded in is the day that HGST reliability will fall.

          My suspicion is that the chinese will not let WD fully merge HGST until SSDs get down to similar pricing/Gb and up to similar capacity as mechanical drives. On current trends that’s about 2-3 years off.

  • Christopher Courtney

    And what do these numbers mean?
    Next to nothing really, due to a very small sample size.

    • ranhalt

      How constructive. Can’t wait to hear back from your research on the matter.

    • That’s part of why we do it. We’re hoping if we release our stats, folks with large farms like ourselves will do the same and we can get some fun anecdata around different use-cases and reliabilities. We’ll keep fighting the good fight though, eventually we’ll have a large sample size…though 40,000+ drives is a pretty good start.

    • seancaldwell

      40,000+ drives is not a very small sample size. You can gather enough data to make a statistical prediction and confidence intervals with a few thousand drives.

      Wish Google or Facebook would release their data on 500,000+ drives, but this is the best large deployment data most people can see. Very thankful Backblaze shares it.