I have a system with

2 x 1TB (NVMe SSDs)
2 x 2TB (SATA SSDs)

disks and would like to create a RAID 1 system using all available disks (so I have a 3TB RAID1 system at the end). Unfortunately, most examples for how to set this up are for just two disks.

What's the recommended approach here?

  • Create RAID1s for each disk size, i.e. mirror the disks of the same kind and then create one large logical volume on top of that?
  • Or is there some other, smarter approach?

If someone has a step by step recipe (or a good link to one), that would be very much appreciated.

Also, do I need a separate system or boot partition?

Thank you.

1 Answer 1


That is heavily depends on what are you aiming at, because NVMe vs SATA SSD speeds & latency have a huge difference.

Personally, I would've created two different sets of LVM volume groups (VG). One for the NVMe and other for SSD and manually assigning them for a different tasks. I.e. NVMe for IO heavy tasks, like DBs and SSD for more generic storage. Of course you can just combine them into a single VG, but that way you are basically "slowing down" NVMe to SATA speed. Well... not really, but almost.

As for the booting - if you have EFI mode system and a modern bootloader (i.e. GRUB2) you'll need a separate small sized (256-512Mb would be fine) partition for the EFI file of FAT32 type. But at least EFI system can boot directly from NVMe and GRUB2 can boot directly from Linux RAID + LVM.

  1. Create 1st partition (i.e. with fdisk) on both the the NVMe drives of about 256-512 Mb size. Set it's type to EFI boot.
  2. Create 2nd partition for the remaining of space for 100% allocation. Set it's type to Linux RAID.
  3. Format each of 1st EFI partition to FAT32 (i.e. mkfs.vfat -F32 /dev/nvme0n1p1).
  4. You can follow the same for SSD drives if you want to make them bootable in case if both NVMe fail or just a single Linux RAID partition for data.
  5. Create a 1st RAID array for NVMe: mdadm --create /dev/md1 -l 1 -n 2 -b internal /dev/nvme0n1p2 /dev/nvme1n1p2.
  6. Create a 2nd RAID array for SSD members: mdadm --create /dev/md2 -l 1 -n 2 /dev/sda2 /dev/sdb2 (if you've created EFI partitions there or just sda1/sdb1 if not).
  7. Create LVM PVs out of newly created arrays: pvcreate /dev/md1 && pvcreate /dev/md2.
  8. Create VGs & LVs on top of the PV. If you still want to combine them, you create 2 PVs and later adding both to same VG.
  9. Make sure to mount EFI partitions & install a proper bootloader on each of the drives. Like this for 1st NVMe drive: mount /dev/nvme0n1p1 /boot/efi && grub-install /dev/nvme0n1.

Note that you can't have a RAID array for EFI partitions. Well... not really, there are some tricks, but I don't think they worth it because there is nothing unrecoverable storage on those. It's just the small binary so that the EFI "bios" can boot your bootloader. Even in case it fails, you can still boot your system of some sort of Live image (like SuperGRUBdisk) and reinstall it with grub-install again.

  • Thank you for the comprehensive answer. You recommend using mdadm for the RAIDs. Is that somehow better than using LVM for the RAID? I.e., is there a reason for this choice?
    – Hendrik
    Oct 5, 2021 at 12:38
  • 3
    LVM are logical volume tool, not RAID. While LVM can create LV mirror, it's still not considered as a full RAID array. It's less flexible and mature than a mdadm in terms of creating mirrored arrays.
    – NStorm
    Oct 5, 2021 at 13:03
  • 2
    Besides, from this question serverfault.com/questions/741103/lvm-raid-1-ssd-caching?rq=1 : "Sidenote: a quick search reveals that LVM mirroring is generally considered to be much slower than Linux software RAID. But I think you can reclaim some of that lost performance with some tuning."
    – NStorm
    Oct 5, 2021 at 13:05
  • 1
    This is the correct answer. Highly recommend @NStorm's opinion about not combining SATA and NVMe SSDs into the same pool. It's very likely that your NVMe devices have bandwidth and IOPS capabilities that are many times what your SATA SSD's are capable of. If you make a single volume from them, your performance will be that of the lowest common denominator. It'd be akin to putting SSD's and spinners in the same pool. Oct 6, 2021 at 13:01
  • On the contrary, it is often desirable to combine them into a single pool. For example, when you configure caching, both cache and the backend volumes must be in the same VG, obviously one of them is some kind of fast SSD and other is slower. Or, you make a thin pool and want to put metadata into faster device, while leaving the data on slower device. In any case, even when PVs are combined into VG, you still can have full control on the actual LV placement, so if you manage them carefully afterwards you won't have any performance penalty, but you'll have flexibility. Oct 6, 2021 at 13:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.