Howdy, Stranger!

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


Do you consider ECC RAM for servers important?
New on LowEndTalk? Please Register and read our Community Rules.

Do you consider ECC RAM for servers important?

HxxxHxxx Member

Been noticing for a little while that dedicated server providers sometimes don't include ECC RAM in their offerings. More prominent with Ryzen servers. Nothing against Ryzen but the server motherboards (i.e. Asrock) for these CPU's usually have support for ECC Unbuffered.

I do consider ECC important and usually try to not rent servers without it. Wondering if nowadays you (as customer) consider ECC RAM at all or don't even consider it?

It could be the case of customers not wanting ECC at all and providers following the behavior.

Feel free to discuss. All opinions welcome.

I attached a poll. Not sure if it will work.

ECC RAM
  1. Do you consider ECC RAM for servers important?134 votes
    1. Yes
      57.46%
    2. No
      42.54%
«1

Comments

  • jarjar Member, Patron Provider
    edited July 18

    I can't attribute enough problems to not having it, to justify going out of my way to have it. Probably takes a really large sample size for the data to stand out.

    Thanked by 2Erisa MikeA
  • cybertechcybertech Member
    edited July 18

    edited wrong comment

  • stefemanstefeman Member

    If its a DB or large data storage, and not a hobby then yes.

    If the data or its integrity actually matters, then ECC is must.

  • luckypenguinluckypenguin Member
    edited July 18

    The probability of a hard drive failure is much higher than a RAM bitflip once a year.
    Or other components, power outage, CPU, that will eventually make you more dependent on
    redundant setups (RAID etc) and regular backups. Can't think of a scenario where ECC would
    br the only thing missing and got you screwed, while all above things were in a good shape.
    I remember Google did a study and came to a conclusion that less than 10% of their
    DIMMs were suspected to have single bitflips in a year cycle.

    Thanked by 1letlover
  • LTnigerLTniger Member

    I never been directly impacted by non-ECC RAM problem. Anything peoduction grade generating dollars? Dump it on big players and forget it. Home stuff - who the fck cares...

  • HxxxHxxx Member

    Interesting results so far. Also the comments. Thanks for participating.

  • tjntjn Member
    edited July 18

    If it’s a production server/cluster that has a DB or ZFS file system, yes.

  • jackbjackb Member, Host Rep

    I'd rate redundant power supplies higher to be honest. Given the choice I'd still use ECC but I'd say redundant power supplies will prevent more outages.

  • darkimmortaldarkimmortal Member
    edited July 18

    I find ECC is more important for when a ram stick fails, than for bitflips. A failed ram stick with many errors, but not enough for the server to fail to boot/function, will hose data in short order.
    I’ve had it happen multiple times and it is enough of a nightmare that I get ECC in everything, including home servers and PCs.
    It’s an insurance policy like RAID - you can restore from backup, but it would be nicer to not need to. I find it more important than RAID, as with disks you know when shit has hit the fan, but not necessarily so with RAM. You might end up backing up corrupt data.
    My anecdotal data is one reported bitflip in 10 years, and 3 data loss events due to bad non-ECC RAM in the same time period

  • comXyzcomXyz Member

    It's all about luck.

    ECC helps correcting single bitflip errors, and based on a research, it happens every 14 to 40 hours per Gigabit of DRAM. The single bitflip errors mostly affect nothing, and usually no ones even notice.

    If you're out of luck, you might even get double bitflip errors, ECC RAM can't deal with those errors, and all of your data might corrupt.

  • aliletalilet Member

    What exactly does ECC do? Let's say I am downloading something on torrent and at exactly 39% for some reason it is unable to write that byte to disk so ECC will try again and non ECC wont?

  • HxxxHxxx Member

    @alilet ECC is more about preventing a memory error from corrupting your data.

    In example a bit flip that turns $50.00 into $5.

    It also helps with server stability overall.

  • I am surprised on the comments here, I wouldn't dream hosting a production server without ECC RAM. I am even starting to use ECC on my zen3 personal desktop.

  • oplinkoplink Member, Patron Provider

    All of our servers ded/vps run ECC memory.

    Thanked by 1Hxxx
  • fadedmaplefadedmaple Member
    edited July 18

    @starservices said:
    I am surprised on the comments here, I wouldn't dream hosting a production server without ECC RAM. I am even starting to use ECC on my zen3 personal desktop.

    Because the problem of 99.99% low end providers is not ECC. Usually it's an oversold bandwidth and CPU, disk corruption, network outage or even misconfiguration. Would anyone pursue ECC when 99.99% SLA are unreachable?

    Thanked by 1martheen
  • HxxxHxxx Member
    edited July 18

    @fadedmaple said:

    @starservices said:
    I am surprised on the comments here, I wouldn't dream hosting a production server without ECC RAM. I am even starting to use ECC on my zen3 personal desktop.

    Because the problem of 99.99% low end providers is not ECC. Usually it's an oversold bandwidth and CPU, disk corruption, network outage or even misconfiguration. Would anyone pursue ECC when 99.99% SLA are unreachable?

    While this is probably the case for many, I'm mostly hosting production and not for fun stuff, therefore is all with medium to large providers.

    So far the poll is been interesting.

    My opinion on this is similar to what @starservices said but I do understand not everyone is hosting apps that works with numbers or mission critical.

    Certainly a server used for torrent or Plex , probably meaningless if data gets corrupted.

  • jsgjsg Member, Resident Benchmarker

    RAM errors happen by far more often than many think (in a typical server every couple of minutes or tens of minutes) and No, one should not shrug them off as insignificant because they can cause a lot of problems.
    So my personal rule of thumb is this: I don't worry about normal everyday servers where even a forced reboot is but an inconvenience. But with anything of any importance like DB servers or anything concerning customer, financial, or other critical or sensitive data I make sure that the system has ECC.
    As a side note, in the low end segment ECC understandably is something that can't reasonably be presumed as granted, so expressly ask your provider before buying.

    Thanked by 1Hxxx
  • Can be ignored if the project is not important. But if data integrity is must then ECC is important.

    Someone I know once gave me access to their server which got crashed. I looked in the error log and found out there is issue on config file on line X. I looked on that line and there was an unknown character while the whole config file was in English. Fixed that and restarted the service.

  • ericlsericls Member, Patron Provider

    For servers that deals with persisted data, yea. For servers that just computes, no

  • letloverletlover Member
    edited July 18

    If ECC is that important for general web hosting, Hetzner would never prosper to become the current big player.

  • letloverletlover Member
    edited July 18

    If hosting health or financial databases, then all kinds of redundancy are needed. postgresql with more data fidelity checking etc are needed at the software level. HA with heartbeat clusters are needed. ECC of course is needed here. But as you can see, ECC for general web hosting using mysql is more an insurance but not critically needed.

  • letloverletlover Member

    As always, it depends.

  • HxxxHxxx Member

    @letlover hetzner do have ECC available for Ryzen dedis. For example.

  • letloverletlover Member

    @Hxxx said:
    @letlover hetzner do have ECC available for Ryzen dedis. For example.

    I understand. But its cheapest dedis are non ECC. And at the beginning Hetzner started with consumer grade electronics servers. Now it grows so big.

    I think ECC is like Xeon cpu. Good to have as insurance, but not essential for operation for general purpose.

  • lonealonea Member, Host Rep

    Yes, it allows you to quickly identify which stick of ram is causing problem.

    No chance with non-ecc

  • HxxxHxxx Member

    Noticed / read that the Intel's new i9 line does support ECC when paired with the proper mobo + RAM ECC. Not sure if anyone here have tried that. It came to me as strange since Intel's usually exclude ECC RAM from the consumer grade CPU's.

  • @Hxxx said:
    Noticed / read that the Intel's new i9 line does support ECC when paired with the proper mobo + RAM ECC. Not sure if anyone here have tried that. It came to me as strange since Intel's usually exclude ECC RAM from the consumer grade CPU's.

    Linus already ranted about it against Intel.
    https://www.theregister.com/2021/01/04/linus_torvalds_intel_killed_ecc
    https://arstechnica.com/gadgets/2021/01/linus-torvalds-blames-intel-for-lack-of-ecc-ram-in-consumer-pcs/

  • ArkasArkas Member

    All my production servers use ECC Ram, all other servers, I don't really care :smile:

    Thanked by 1Hxxx
  • MechanicWebMechanicWeb Member, Patron Provider

    I think this question is wrong. The preference for ECC RAM is undeniable.

    It is also important to note two things.

    If you can use ECC, do it.

    Every error in memory does not result in data corruption. The chances of data corruption from memory errors are slim. You are more likely to observe a server crash due to software, human error, and/or other hardware issues than due to non-ECC RAM.

    Having said that, I am not aware of any server crashes or data corruption issues due to non ECC RAM in recent years. But I would love to know that someone had that experience.

    Thanked by 1Hxxx
  • HxxxHxxx Member
    edited July 26

    Is there any provider in these parts providing shared hosting or VM's currently not using ECC RAM and still having good results?

    Would love to know your experience. Just curious to know if you had seen any significant errors, even thought much of these would go undetected which is the issue.

Sign In or Register to comment.