4 ms·

An equally interesting question would be... "Are the SSDs you are likely to buy more reliable than the HDDs you are likely to buy" ?

xnx 2y ago
For your usage patterns
nine_k 2y ago
Even more generally, plot the reliability of HDD and SSD along the price axis. Plot such charts for a few widespread sizes (say, 256 GiB to 8 TiB). For bonus points, plot SLC, DLC, QLC flash drives separately, and normal / shingled HDDs separately. Alternatively, plot price per GB for given levels of reliability. I suspect that this chart is going to have distinct sweet spots.
- snailmailstare 2y ago
  That's a lot of required purchases of the wrong hardware for the business that the company providing the statistics is in.
  - do_not_redeem 2y ago
    Yep. I'm confident the large cloud providers (Google, Amazon, etc) run enough SSDs that it's worth them having this data internally, but there's probably no motivation to share it, or maybe they see having the data as some sort of competitive advantage.
    - homebrewer 2y ago
      FWIW, Josef Bacik, one of the main developers of btrfs and a long-time Facebook employee, said many times that most Facebook servers use the crappiest SSDs they can buy. I won't be delving into hour-long searches through the mailing lists to support a casual comment, but here's something I could dig out quickly: https://lwn.net/Articles/824855/ https://lwn.net/Articles/824855/
      - braiamp 2y ago
        > I won't be delving into hour-long searches through the mailing lists to support a casual comment Don't worry fam, I got you. https://lwn.net/ml/fedora-devel/03fbbb9a-7e74-fc49-c663-32722d6f75a2@toxicpanda.com/ https://lwn.net/ml/fedora-devel/03fbbb9a-7e74-fc49-c663-3272... Source via: https://news.ycombinator.com/item?id=30443761 https://news.ycombinator.com/item?id=30443761
        devsda 2y ago
        Not sure what the take away is from their experience. Is it 1. Consumer grade/crappy SSDs are still reliable enough at scale and rarely fail that it is easier to manage occasional failures ? 2. The cost of engineering effort to keep things together with crappy ssds failing often is less compared to the expensive enterprise grade hardware ? Or is it something else ?
        pjc50 2y ago
        Once you build a distributed system that's large, it has to tolerate routine node failures. Once you have a fault tolerant system, and some idea of failure rates, you can then ask "is it worth having X units of less reliable hardware or Y units of more reliable hardware? How many will still be running after a year?" You then find out that buying a few more cheap nodes compensates completely for the lower reliability.
        nine_k 2y ago
        One more step, and it will conjoin with Erlang's "Let it crash" approach.
        snailmailstare 2y ago
        This thread seems overly complex given that an engineer tasked with designing a RAID storage to replace an unplanned haberdashery would probably start with looking up what the requested acronym stands for.
- throwaway48476 2y ago
  Idle spinning hard drives consumes multiple times that of idle SSDs. Over the life of the drive this becomes significant as well.
  - justinclift 2y ago
    > idle SSDs Bearing in mind that SSDs used in data centres (at least used to) not have much in the way of low power idlying mode like consumer grade SSDs do (due to laptop requirements).
    - throwaway48476 2y ago
      Data center SSDs quote idle power much lower than active power, and lower than HDD too. Data centers care plenty about power consumption.
      - justinclift 2y ago
        Good. The stuff I still deal with (older generation SAS2/3 gear) isn't that way. ;)
crabbone 2y ago
Here, adding more axes to consider: In enterprise settings, what you are likely to be interested in is how easy it is to recover a RAID with a failing member. Now, here's an interesting part: the larger the disk participating in the RAID, the more is recovery susceptible to failures of the disk controller. It's possible that the failure rate on the controller is so high that whenever you are trying to recover the data, you will have a very high (>50%) chance of introducing an error... and so a RAID virtually becomes unrecoverable. SATA controllers are in that later category (and that's partially why enterprise users preferred SAS before NVMe + PCIe became a thing).

xnx 2y ago

For your usage patterns

nine_k 2y ago

Even more generally, plot the reliability of HDD and SSD along the price axis. Plot such charts for a few widespread sizes (say, 256 GiB to 8 TiB). For bonus points, plot SLC, DLC, QLC flash drives separately, and normal / shingled HDDs separately. Alternatively, plot price per GB for given levels of reliability. I suspect that this chart is going to have distinct sweet spots.

snailmailstare 2y ago
That's a lot of required purchases of the wrong hardware for the business that the company providing the statistics is in.
- do_not_redeem 2y ago
  Yep. I'm confident the large cloud providers (Google, Amazon, etc) run enough SSDs that it's worth them having this data internally, but there's probably no motivation to share it, or maybe they see having the data as some sort of competitive advantage.
  - homebrewer 2y ago
    FWIW, Josef Bacik, one of the main developers of btrfs and a long-time Facebook employee, said many times that most Facebook servers use the crappiest SSDs they can buy. I won't be delving into hour-long searches through the mailing lists to support a casual comment, but here's something I could dig out quickly: https://lwn.net/Articles/824855/ https://lwn.net/Articles/824855/
    - braiamp 2y ago
      > I won't be delving into hour-long searches through the mailing lists to support a casual comment Don't worry fam, I got you. https://lwn.net/ml/fedora-devel/03fbbb9a-7e74-fc49-c663-32722d6f75a2@toxicpanda.com/ https://lwn.net/ml/fedora-devel/03fbbb9a-7e74-fc49-c663-3272... Source via: https://news.ycombinator.com/item?id=30443761 https://news.ycombinator.com/item?id=30443761
      - devsda 2y ago
        Not sure what the take away is from their experience. Is it 1. Consumer grade/crappy SSDs are still reliable enough at scale and rarely fail that it is easier to manage occasional failures ? 2. The cost of engineering effort to keep things together with crappy ssds failing often is less compared to the expensive enterprise grade hardware ? Or is it something else ?
        pjc50 2y ago
        Once you build a distributed system that's large, it has to tolerate routine node failures. Once you have a fault tolerant system, and some idea of failure rates, you can then ask "is it worth having X units of less reliable hardware or Y units of more reliable hardware? How many will still be running after a year?" You then find out that buying a few more cheap nodes compensates completely for the lower reliability.
        nine_k 2y ago
        One more step, and it will conjoin with Erlang's "Let it crash" approach.
        snailmailstare 2y ago
        This thread seems overly complex given that an engineer tasked with designing a RAID storage to replace an unplanned haberdashery would probably start with looking up what the requested acronym stands for.
throwaway48476 2y ago
Idle spinning hard drives consumes multiple times that of idle SSDs. Over the life of the drive this becomes significant as well.
- justinclift 2y ago
  > idle SSDs Bearing in mind that SSDs used in data centres (at least used to) not have much in the way of low power idlying mode like consumer grade SSDs do (due to laptop requirements).
  - throwaway48476 2y ago
    Data center SSDs quote idle power much lower than active power, and lower than HDD too. Data centers care plenty about power consumption.
    - justinclift 2y ago
      Good. The stuff I still deal with (older generation SAS2/3 gear) isn't that way. ;)

crabbone 2y ago

Here, adding more axes to consider: In enterprise settings, what you are likely to be interested in is how easy it is to recover a RAID with a failing member. Now, here's an interesting part: the larger the disk participating in the RAID, the more is recovery susceptible to failures of the disk controller. It's possible that the failure rate on the controller is so high that whenever you are trying to recover the data, you will have a very high (>50%) chance of introducing an error... and so a RAID virtually becomes unrecoverable. SATA controllers are in that later category (and that's partially why enterprise users preferred SAS before NVMe + PCIe became a thing).