Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!


Does raid-5 drives last soon before raid-10 drives?
New on LowEndTalk? Please Register and read our Community Rules.

All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.

Does raid-5 drives last soon before raid-10 drives?

Hello,

I know it is some thing strange to ask, but I was in discussion today with IT guy in one of the data centers, he said the raid-5 affect the drive health soon, as it continues to read + write which affect the drive to be last or died before it is aged! And it is better to used Raid-10.
From my perspective, the rebuilding in Raid-5 is taking much longer than Raid-10 and drive failures recovery would be better.

Any thoughts?

Regards

Comments

  • @alshahad said: From my perspective, the rebuilding in Raid-5 is taking much longer than Raid-10 and drive failures recovery would be better.

    better than me nobody know what raid5 means , definetly go with raid 10 , almost 99% procent rebuild will faill if have huge quantity of hdd and big data hdd.

    Thanked by 2raindog308 pike
  • Like all raid setups depends what your going for. Raid 10 is generally a solid balance however if you want more redundancy raid 6 is a better option.

  • dfroedfroe Member, Host Rep

    RAID 5 means you only have n+1 redundancy. So if one drive fails, all others drives must be in perfect condition during the whole rebuild process. It's a question of probability whether this risk is acceptable for you or not.

    RAID 6 will give you n+2 redundancy which means you can tolerate one other failure in your array during rebuild.

    In a RAID 10 setup you have a bunch of mirrors/pairs. If one drive fails only the other mirrored partner drive needs to be in perfect shape. And of course rebuild is faster since only one pair of drives must be rebuild.

    It all depends on your use case and how important data availability is for you. In the end this is just math and calculating with probabilities. Keep in mind that RAID is never a backup so you must backup your data somewhere in every case.

  • cociucociu Member
    edited November 2020

    @dfroe said: RAID 5 means you only have n+1 redundancy. So if one drive fails, all others drives must be in perfect condition during the whole rebuild process. It's a question of probability whether this risk is acceptable for you or not.

    if your data is important please dont raid5 , finnaly you will lost all all case what i have in a raid 5 between 14x hdd of 12 tb rebuild was faill, to note in most of the case the hdd was less 6 month old ... so for me raid5 is not exist anymore.

    i telling a guy ho have xx petabytes here like storage (not to make a big person or have any kind of publicity but i hate raid5 in this moment because like i told you the rebuild never was pass )

  • dfroedfroe Member, Host Rep

    @cociu said: finnaly you will lost all all case what i have in a raid 5 between 14x hdd of 12 tb rebuild was faill

    Consumer drives are rated with an error bit rate of 10^14 which means statistically you'll get a read error every 12.5 TB. Enterprise drives are rated 10^15 resulting in a read error every 125 TB.

    Putting 14x 12 TB = 168 TB into one array with only one piece of parity means 156 TB must be perfectly readable without any error during rebuild. This won't end well and sounds quite risky even with enterprise grade drives. :)

    For such huge arrays I would consider more parity or reducing the drives per array - which is basically what RAID 10 does.

    Thanked by 1TimboJones
  • @dfroe said: dfroe

    totally true , unfortunate i have see this with my eyes .... so we don`t use it anymore.

  • jsgjsg Member, Resident Benchmarker

    @dfroe said:
    RAID 5 means you only have n+1 redundancy ...
    RAID 6 will give you n+2 redundancy ...
    In a RAID 10 setup you have a bunch of mirrors/pairs. If one drive fails only the other mirrored partner drive needs to be in perfect shape. And of course rebuild is faster since only one pair of drives must be rebuild.

    Yes, but that's just 1 point. The other point - and the one that drives many towards Raid 5 or 6 - is space efficiency, which usually is much better with Raid 5 or 6.

    What people must understand is that the whole Raid issue often is more complex than many seem to think and that there is no generally good rule

    It all depends on your use case and how important data availability is for you. In the end this is just math and calculating with probabilities. Keep in mind that RAID is never a backup so you must backup your data somewhere in every case.

    Yes.

  • @cociu said: in a raid 5 between 14x hdd of 12 tb

    please tell you are joking and did not run such a setup ever? I mean that this is going to fail, is not something newly discovered in the last two or three years but quite known for a while...

  • @Falzo said: please tell you are joking and did not run such a setup ever? I mean that this is going to fail, is not something newly discovered in the last two or three years but quite known for a while...

    yes i have been this idiot , lesson learned ...

  • raindog308raindog308 Administrator, Veteran

    @dfroe said: Consumer drives are rated with an error bit rate of 10^14 which means statistically you'll get a read error every 12.5 TB.

    Hold up...if I have a 6TB drive, that means on average after it's been read through twice I will get a read error?

    My home arrays are sets of 2x6TB in RAID-1. Every backup reads through them and I don't see anything remotely like that...years go by without a read error.

  • dfroedfroe Member, Host Rep

    @raindog308 said: Hold up...if I have a 6TB drive, that means on average after it's been read through twice I will get a read error?

    Statistically from what is stated in the datasheet by the vendor, yes. But strictly speaking it should be worst case, not average.
    For example WD specifies its WD RED including the 14 TB model with <1 non-recoverable error per 10^14 bits read.
    Of course it says "less than", but vice versa you shouldn't expect being able to read the whole drive without any bit error.
    So even the HDD vendor doesn't guarantee the whole drive can be read without an error. Which might be surprising.

    From my experience on ~15 TB storage pools over the last 10+ years bit errors are very rare but they happen. I remember a few times when ZFS reported (and corrected) wrong blocks on one of the drives during a scrub checksum check.

    This is also why I love ZFS, not only for its ease of management (if you are familiar with it) but also the fact that everything is checksummed and verified upon read. Of course you should also consider ECC memory as otherwise corrupt data may end up on all your mirrored drives.

    In real life the error rate might be more like 10^15 (once every 125 TB) so it takes a lot of time for an ordinary user to read that bunch of data. And if it's not checksummed or otherwise verified, one probably doesn't even notice a wrong bit within a large movie stream.

  • FalzoFalzo Member
    edited November 2020

    @raindog308 afaik there is a distinction between 'just' read errors and unrecoverable read errors / UREs. the latter describe a point where you hit a read error (happens more often) but the electronics of the hard disc can't recover from that. only then a read will fail and the 10^14 or 10^15 are averages about that probability. it is not said that it must happen ;-)

    also in a healthy raid-1 you would not have a problem anyway, because the data can be read from the mirror, so it most likely would not even happen that the potentially faulty discs tries to reread after an error mutliple times etc.
    issues only occur during a rebuild, where all data must be read successfully to verify integrity and rewrite missing parity (or mirror) and there is no other drive anymore to get the missing data from.

    for the backups running, apart from the mentioned availability of the data from two disks I would also think, that the reads might not be as much as you think, at least if you do not run full backups all the time - could be your backup is more likely just running a stat against the files to compare and pick new ones, no?

    I've also read about the famous western digital red somewhere, that these silently might return dummy data on an actual URE to not have the array die, but no idea if that's true or just an urban myth. probably everything within these discussions has to be taken with a grain of salt.

    good reads: https://serverfault.com/a/812929 , https://serverfault.com/questions/889668/what-will-happen-if-encountered-a-ure

    it's probably a gamble on probabilities nevertheless... with very large disks I'd say you're better off running them as JBOD anyway, because with an URE then you might just loose a file or two but won't get stuck with rebuilds and fscks ;-)

    Thanked by 3saibal raindog308 dfroe
  • jsgjsg Member, Resident Benchmarker

    @raindog308 said:

    @dfroe said: Consumer drives are rated with an error bit rate of 10^14 which means statistically you'll get a read error every 12.5 TB.

    Hold up...if I have a 6TB drive, that means on average after it's been read through twice I will get a read error?

    My home arrays are sets of 2x6TB in RAID-1. Every backup reads through them and I don't see anything remotely like that...years go by without a read error.

    You are right, @dfroe got it wrong.

    1 TiB is 1024 * 1024 * 1024 * 1024 bytes which is a bit less than 1.1 TB (A TB being 10 to the 12 bytes and a TiB being 2 to the 40 bytes).

    1 TB is 10 to the 12 and hence 10 to the 14 are 100 TB or about 90 TiB.

    But, there dfroe is correct, a URE is unrecoverable as seen by the drives controller and there may be layers on top, e.g. ZFS, which have their own checksums or even erasure codes, plus even a simple Raid 1 controller is usually smart enough to chose the data from the good drives sector (and copy it from there to the "bad" drive) so you may indeed not have seen an error when reading from your 6 TB drive for some years.

    That said, one should generally not simply trust in hardware, as even ECC memory is limited wrt how man errors it can detect and correct, so one should always have up to date backups (and do high reliability computations on at least 3 systems). Funny btw that many people know that disks have a rather limited reliability but very few know that RAM, even ECC RAM has limited reliability, and considering the amount of data pushed to and pulled from memory more consideration seems justified).

  • dfroedfroe Member, Host Rep

    @jsg said: hence 10 to the 14 are 100 TB or about 90 TiB

    Just to double check, are we talking about bytes or bits?
    According to the vendor datasheets it's (less than) 1 non-recoverable error per 10^14 bits (not bytes). So we may hit one error every 10^14 bits (which is 12.5 TB).

    Thanked by 2Falzo TimboJones
  • @dfroe said: Just to double check, are we talking about bytes or bits?

    According to the vendor datasheets it's (less than) 1 non-recoverable error per 10^14 bits (not bytes). So we may hit one error every 10^14 bits (which is 12.5 TB).

    2 matematic here ... so raid10 will not faill and not head cake ...

  • dfroedfroe Member, Host Rep

    @cociu said: so raid10 will not faill

    Well, the end is neigh either way and shit happens all the time, there's no doubt about that. :D

  • jsgjsg Member, Resident Benchmarker

    @dfroe said:

    @jsg said: hence 10 to the 14 are 100 TB or about 90 TiB

    Just to double check, are we talking about bytes or bits?
    According to the vendor datasheets it's (less than) 1 non-recoverable error per 10^14 bits (not bytes). So we may hit one error every 10^14 bits (which is 12.5 TB).

    If the spec is indeed in bits then ca. 12.5 TB is correct and you were right.

  • dfroedfroe Member, Host Rep

    Of course I'm not about who's right or wrong.
    BTW, here's the WD RED datasheet for reference.

    I mainly find it interesting "how faulty" hard disks can be.
    I mean, a 14 TB HDD with an error rate of up to 1/12.5 TB, isn't that like buying a car that breaks down up to once on each tank of fuel? :)

  • cociucociu Member
    edited November 2020

    @dfroe said: isn't that like buying a car that breaks down up to once on each tank of fuel?

    lol , you make my night , my first car (second one) for coincidence was from your country ... and i have drive like 15 km and the motor was exploding , one piston engine was out ... is a real story ... so finnaly i have to go like 7 km to the first gas station to call a taxi because i even dont have credit to my mobile ... what a nice period ...

    so conclusion my first car had less life than one hdd from 2020 edition hahahahah

    Thanked by 3dfroe maverickp lentro
  • jsgjsg Member, Resident Benchmarker
    edited November 2020

    @dfroe said:
    Of course I'm not about who's right or wrong.
    BTW, here's the WD RED datasheet for reference.

    I mainly find it interesting "how faulty" hard disks can be.
    I mean, a 14 TB HDD with an error rate of up to 1/12.5 TB, isn't that like buying a car that breaks down up to once on each tank of fuel? :)

    I guess the driver behind the reliability issues with drives is that demand for capacity grew too much too quickly. Such the drive electronics (and possibly mechanics) got into a situation that's far beyond what they can handle.
    A reasonable life time and the related total read and write volume were within bounds up to something like 10 - 50 GB drives, beyond that the total read/write volume weren't clearly within a healthy fraction of 90% of the drives URE numbers, and nowadays with terabyte drives not just the total volume but increasingly even a single full drive read/write is beyond what the technology can handle reliably.

    There would be a solution, flash based drives, but the desire to save costs leads people to buy hard drives, and the manufacturers just happily go along although they know their URE is not increasing along with ever higher capacity.

    To be honest, I myself sometimes use spindles but at least I stick to my own rules:

    • Do not buy the largest cap. spindles but stay reasonable and use more drives of smaller sizes.
    • Use Raid. Yes, even Raid 6 with (good quality) spindles of <= 3 TB.
    • Use a file system with error detection (checksums, parity)
    • Where sensible use erasure codes on top of that and/or
    • mirror critical data over multiple file systems (as opposed to just drives or Raid arrays)
    • Make backups. Often. Preferably to 3 or at least 2 backup systems.
    • Stay away from WD. They lied and cheated. Why should I assume they won't do it again? Nope. no WD.

    The good news: I guess in a few years (2 or 3 I guess) flash based drives will (a) offer larger cap (in standard packages like 2.5"), and (b) be cheaper or at least only slightly more expensive that spindles.

Sign In or Register to comment.