1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

Latest BackBlaze drive stats

Discussion in 'Storage Drives' started by rotor, 17 Feb 2016.

  1. rotor

    Wise Guy

    Joined: 18 Sep 2012

    Posts: 2,196

  2. drakioned

    Wise Guy

    Joined: 11 Dec 2008

    Posts: 1,186

    Location: London

    Interesting Toshiba does okay despite having basically the cheapest drives.
     
  3. KIA

    Man of Honour

    Joined: 14 Nov 2004

    Posts: 13,577

    Consumer drives in an enterprise environment...

    I'm sceptical.
     
  4. panyan

    Sgarrista

    Joined: 13 Mar 2011

    Posts: 7,522

    Location: Bada Bing

    do we? I missed that memo
     
  5. rotor

    Wise Guy

    Joined: 18 Sep 2012

    Posts: 2,196

    They ordered 16,000 4TB drives in 2015. Says it all.

    Enterprise drives cost what they cost because the vendors know they can get away with it, not because the price in any way reflects the true cost. The manufacturing process is exactly the same.

    We pay over £1,000 for 300GB SSDs from EMC. Are they made with pixie gold fairy unicorn dust? That's over 10x the cost of equivalent (probably faster) consumer SSDs. You could put up with a LOT of failures (assuming your infrastructure is capable of handling the failures, which is what BackBlaze does) for a 90% savings.

    Horses for courses. It's all about risk and liability.
     
  6. rotor

    Wise Guy

    Joined: 18 Sep 2012

    Posts: 2,196

    That's my overall impression in this forum, yes. =)
     
  7. AbsenceJam

    Mobster

    Joined: 2 Nov 2007

    Posts: 4,304

    BackBlaze's results are really meaningless outside of their situation (constant reads/writes 24/7), and complete meaningless in some instances due to the tiny sample sizes.

    Hard drives are mechanical and they can, and do, fail. Rarely there's a bad model - not manufacturer - that's more likely to fail. Backup, then you won't lose anything important and develop an irrational prejudice.
     
  8. Richdog

    Caporegime

    Joined: 8 Sep 2005

    Posts: 26,293

    Location: Utopia

    What do you mean tiny sample sizes?
     
  9. Rroff

    Man of Honour

    Joined: 13 Oct 2006

    Posts: 74,103

    Depends a bit but often enterprise drives come with a different warranty/service agreement to consumer drives and in some cases use uprated components to provide significantly increased MTBFs.

    Depending on requirements you may find they are significantly better workload optimised as well - while maybe not having the raw speeds of a high end consumer drive a lot higher IOPs for instance.

    EDIT: Many years ago I tried using normal consumer HDDs in a server that was parsing and logging significant amounts of data in realtime (probably the equivalent of a few days worth of posts on these forums every hour) and after a few weeks they'd fail - with one literally almost melting :p quite dramatic heat failure under constant load - replaced with an enterprise level drive it ran happily for years.
     
    Last edited: 17 Feb 2016
  10. AbsenceJam

    Mobster

    Joined: 2 Nov 2007

    Posts: 4,304

    They don't even have 50 units of some models in use and they publish data. Some of their previous 'statistics' have included models where they have had around 10 in use.
     
  11. rotor

    Wise Guy

    Joined: 18 Sep 2012

    Posts: 2,196

    They clearly state the numbers of drives in each sample size.
     
  12. AbsenceJam

    Mobster

    Joined: 2 Nov 2007

    Posts: 4,304

    So what is your point?
     
  13. rotor

    Wise Guy

    Joined: 18 Sep 2012

    Posts: 2,196

    That they are providing raw data, which is better than none at all. And they are being up front about it, and have no agenda. Each person is then free to take his or her own conclusions from said data.

    My point is that your use of 'quotes' was meant to imply scorn or mistrust. I don't see why that is warranted.
     
  14. Minstadave

    Caporegime

    Joined: 8 Jan 2004

    Posts: 28,086

    Location: Rutland

    ST3000DM001 - ultimate lemon of a drive, and it's my only mechanical drive in my main rig :(

    Yes Backblazes data is from a different environment to home use but unless you want to survey thousands of home users over years it's about as good as you're going to get realistically.
     
    Last edited: 17 Feb 2016
  15. AbsenceJam

    Mobster

    Joined: 2 Nov 2007

    Posts: 4,304

    Incomplete raw data is provided (nothing on environment, role, etc which I'd definitely want to know), but they aren’t only providing that. They use the statistically insignificant data to draw conclusions and include them in their representations of the data. Not to mention that they know very well what happens when they do this. It’s mostly read and republished as manufacturer X is bad. No agenda? A TON of attention does them no harm…

    And what does your opening post say?
     
  16. AbsenceJam

    Mobster

    Joined: 2 Nov 2007

    Posts: 4,304

    Is there anything definitive written about this?

    Time was an enterprise drive was usually a physically a different beast with higher IOPs (higher spindle speed) and lower STR (less platter density). But the ones that look the same on paper, is there actually a physical difference? I have never seen anything that conclusively says they use better components or they’re tested and binned.

    Sure they state a higher MTBF and have a longer warranty, but is that legitimate and not just part of the premium they charge?

    They may have an SAS controller, and if SATA different firmware more suited to array use including power management - which could make a difference if you’re trying to use them 24/7?

    External disks, which BackBlaze had/have thousands upon thousands of (removed from their cases), are rumoured to be the worst drives with the least warranty and MTBF, and perhaps the firmware making them even less suitable still.
     
  17. rotor

    Wise Guy

    Joined: 18 Sep 2012

    Posts: 2,196

    No agenda as in not pushing any specific result. Of course they get attention, which is good for them, but I don't see how that taints the results? As far as I'm concerned, the results can speak for themselves.

    What environment? Let's assume it isn't an open tent in the middle of a sandstorm in the Sahara; i.e. environment = "a generic datacenter". What role? Storing people's stuff. I would say 99% of the time these disks are idle, as my logical deduction is that stuff gets backed up, and then extremely rarely touched again. So once disks fill up, they will still receive random access, but on a fairly rare basis.

    My opening post was meant to be facetious, apologies if that didn't come through.

    I guess what I don't get is this: you imply that you would prefer if they didn't release this data at all?
     
  18. lude1962

    Wise Guy

    Joined: 27 Feb 2014

    Posts: 2,152

    more info on their storage pods here.
    A few things stands out they use EVGA Supernova psu's, and 'We have multiple (10-30) pods online at any time accepting data'.
    Looking at the pics they are running the drives vertically, not ideal for a desktop drive.
     
  19. bledd

    Don

    Joined: 21 Oct 2002

    Posts: 46,780

    Location: Parts Unknown

    The underlying rule is..

    Never trust a hard drive

    Keep your data on at least 2 drives. 3 if possible.


    All manufacturers get the odd blip where they have a run of bad drives. Remeber the IBM DeskStar ? Massive failure rate!
     
  20. Mercutio

    Soldato

    Joined: 11 Jun 2003

    Posts: 5,004

    Location: Sheffield, UK

    the backblaze data, while not totally realistic for home use IS using all the drives in the same way. They're getting tortured (to death) so it DOES give a reasonable sample of the overall reliability of a drive in a worst case usage setup.

    I don't understand the mentality of folks that immediately go "but it's not like home use!" and "The data is irrelevant". For the first point, no, it's not... but it IS using consumer drives in a really harsh way and showing which can take the beating the best(?). That to me would imply reliability.
    For the second point (irrelevant data) MTBF (mean times between failure) is basically a measure of how long a drive should typically last in use. Drives that have consistently higher failure rates in the backblaze data will generally show ones that LIKELY have lower MTBF averages and are (overall) less reliable.
     
    Last edited: 19 Feb 2016