Raid 10 boot - how to?

    • Raid 10 boot - how to?

      Im wondering if anyone has tried or is currently doing this. I am looking to copy over my existing single disk ssd boot drive to a raid 10 software (md) raid setup. However I have been testing in VM's with no luck.

      So far I have tried things like clonezilla all the way to just dd'ing entire disk. Problems I have run into so far are:

      -grub gets weird, doesnt want to work with raid 10 specifically
      -dd doesnt seem to partition correctly.
      -clonezilla doesnt actually support restore to md raid.
      -md raid doesnt appear to start in time so that grub can see it.

      Things that I have not tried yet:
      -switching bios from AHCI to RAID (in VM environments this option doesnt exist so I have not got to test yet).
      -trying on actual hardware.

      Any help is super appreciated. I also did a search across the forums and I couldnt find anyone trying raid 10 ... found tons of raid 1 and 0 though.

      OR if someone has a awesome guide to ZFS boot im totally open to that too (I know Proxmox has a native option for this and its Debian based as well).

      Thanks in advance.
    • Re,

      why would you make a RAID10-bootdisk?

      Setting up RAID-devices as boot-devices is more difficult, cause you have to add the raid-modules to the initramfs ... but if you have already a SSD as a boot-device, it's not worth the time ...

      And you have to do this 1st ... bevor installing OMV, so you have to use the Debian-NetInstall-ISO and after all is finished, install OMV over that.

      For making a RAID1 as boot-device, just google arround ... (e.g. edoceo.com/howto/mdadm-raid1)

      Sc0rp
    • Hi,
      Thanks for the response. Dang I was afraid I was missing something with raid setup post-os install.

      I use a SSD as a boot device and its not the best. Mainly because of reliability. A single disk boot disk is problematic for me. I would like performance gains (im seeing a lot of I/O wait) and since my last SSD lasted 11 months (NAND writes were super high - drive died of natural causes, not defective) I would like to put some redundancy in it. Since it looks like raid 10 is out; do you have any experience with ZFS boot post-OS-install?

      Thanks again.

      edit: pre -> post

      edit2: im thinking ZFS because I have used ZFS as boot on Proxmox and its absolutely amazing.
    • Not data drive, im assuming rclone. Havent really looked into the specifics much... isnt it pretty normal to have high read and writes? unless the OS is loaded into ram like ESXi.

      edit: if not raid. What are others doing for when OS drive dies?

      edit2: rclone is like rsync but for backing up to things like S3 or Backblaze B2, etc.

      The post was edited 1 time, last by Lazurixx ().

    • Re,

      Lazurixx wrote:

      isnt it pretty normal to have high read and writes?
      No, that's not normal for system-drive ... in case of the writes. May be you have to less RAM, or you have misconfigured a "cache system" ...

      If you have lot's of writes on the SSD, you should use a pro-grade SSD anyway (like Samsung Pro series) ... but you should check, which configuration causes the high write-load, because that will destroy any SSD - and affects of course ALL members of an RAID1! (and, for RAID1, it may double the read-speed and read-I/O's, but the write-speed and the write I/O's remains the same!).

      Lazurixx wrote:

      unless the OS is loaded into ram like ESXi.
      Since ESXi is also based on a unix/linux core, the behavior ist the same for all "OS" this days ... read the files from disk and buffer as much of them in RAM (Cache-Area) as there is space ... the rest is used for file-caching (FS related).

      Nibb31 wrote:

      If the OS dies, you reinstall. It's usually faster than messing around with clonezilla.
      That depends on the configuration of the whole NAS ... CloneZilla is a bare metal backup incl. network (server/client), so U can use another host as server and other possibilities ...

      RAID1 as boot-device brings only more reliability if different drives of the same technology are used (e.g. one HDD from Seagate and one from WD - of course with the same base specs), on SSDs i didn't recommend that, there is a good backup-strategy better (rsync to a backup-dir, Image, internal backup-plugin, ...)

      Sc0rp
    • Sc0rp wrote:

      RAID1 as boot-device brings only more reliability if different drives of the same technology are used (e.g. one HDD from Seagate and one from WD - of course with the same base specs), on SSDs i didn't recommend that
      With RAID-1 and SSDs not following this rule and using exactly the same devices (especially when even firmware revision is the same) is really dangerous since SSDs die for other reasons than spinning rust (eg. firmware failures, we had this with almost all vendors in the past that there were bugs happening after n hours of operation --> with same type of SSDs configured as RAID-1 such bugs will render the whole array unusable within a very short time then).

      Also we should not forget that RAID-1 is pretty useless anyway, that as you already pointed out RAID-10 speed improvements are only with reads (so nothing to win here with OMV other than boot time decreasing by 2 nanoseconds), that if there are constant writes this needs a quick check (using opensnoop for example) and that with flash media write amplification should be considered the real problem.

      The same GByte of data written one byte at a time to an SSD, SD card or pendrive will wear out the media 1000 times faster compared to writing 1KB at a time. That's what OMV's flashmemory plugin is for and that's why people want to disable monitoring in such situations (since RRD databases are written continously).

      If I would like to have some fancy 'OS drive' improvements I would choose btrfs, make regular snapshots prior to 'apt upgrade', send those snapshots to another disk/host to have some sort of backup, do regular scrubs to see whether data got corrupted or not (it's almost 2018 now, this is all possible and even works flawlessly. All the OMV images for toys -- ARM devices like Banana Pi -- are prepared for this)

      And if I would fear my 'OS drive' dying I would transform my btrfs OS partition into btrfs's own RAID-1 (not mdraid of course) to get even self-healing capabilities to be protected against (a) single drive fail(ures)
    • Sc0rp wrote:

      No, that's not normal for system-drive ... in case of the writes. May be you have to less RAM, or you have misconfigured a "cache system" ...
      The installation I currently have setup is just the normal OMV 3 installation media. I do have low ram for what operations I have going on - but I have been paying attention to swap and there has not been rampant changes.

      Is there a good way to check allocation of "cache systems"?

      tkaiser wrote:

      With RAID-1 and SSDs not following this rule and using exactly the same devices (especially when even firmware revision is the same) is really dangerous since SSDs die for other reasons than spinning rust (eg. firmware failures, we had this with almost all vendors in the past that there were bugs happening after n hours of operation --> with same type of SSDs configured as RAID-1 such bugs will render the whole array unusable within a very short time then).
      Interesting you mention this - I too have had random PCB/firmware related issues with some SSD's (specifically Crucial SSD's so far).

      Any good way to avoid issues with the boot disk that I am not thinking of (besides Raid)?

      tkaiser wrote:

      The same GByte of data written one byte at a time to an SSD, SD card or pendrive will wear out the media 1000 times faster compared to writing 1KB at a time. That's what OMV's flashmemory plugin is for and that's why people want to disable monitoring in such situations (since RRD databases are written continuously).
      Ok I am curious here - I have used the flashmem plugin for USB sticks and it worked well. I wasnt sure about its use case with SSD's. Will this create a significant amount of overhead in ram though? It's basically everthing that would have gone to swap, right?


      Thanks again to everyone that has posted - I appreciate the help!
    • Lazurixx wrote:

      Any good way to avoid issues with the boot disk that I am not thinking of (besides Raid)?
      Forget about RAID, better backup your boot disk regularly if you are concernced about it failing (for whatever reasons -- my OMV boxes boot from SD cards and I'm not concerned at all). How should RAID help here? If your boot config gets corrupted it's corrupted on two disks at the same time.

      And flashmemory plugin is not related to swap at all (quite the opposite). Swap is using disk storage if RAM gets low while with the flashmemory plugin we use free RAM to buffer stuff that should go to disk (so yeah, if now a power failure happens these changes are lost). The flashmemory plugin uses tmpfs to collect changes in RAM that then get written every hour to disk in a single step. And this (the single step) greatly reduces write amplification and is reponsible for your flash media wearing out ten to hundred times later.
    • Users Online 1

      1 Guest