Suppose you have several drives of different geometries which you would like to join together into a large storage pool. There are several options available:

• RAID-ing the devices together, but that may incur a loss in storage space or storage reliability.
• JBOD-ing the devices, but that will mean that if one drive fails, then all the data is lost, similar to stripe RAID0.

A different option is to use mhddfs which is a filesystem in user space (fuse) that can merge several large filesystems into a single large filesystem. Since mhddfs is a filesystem in user-space, it does not even require root privileges and also does not require the drives to be partitioned using the same filesystem.

Moreover, the advantage is that if one drive fails, then only the data on that drive will be lost while the other drives will be mountable, readable and contain data you can access.

In order to use mhddfs, it must be installed. On a Debian Linux system, this can be achieved with:

aptitude install mhddfs

which is all that is required.

The example setup presented here consists of two drives:

• /dev/sdb1 of 1TB
• /dev/sdc1 of 500MB

In order to merge these two drives together, we first need a filesystem. In our case, we use ext4 for both drives - although we could have picked something different for each, and issue the commands:

mkfs.ext4 -O 64bit,has_journal,extents,huge_file,flex_bg,uninit_bg,dir_nlink,extra_isize,dir_index -i 4194304 /dev/sdb1

respectively:

mkfs.ext4 -O 64bit,has_journal,extents,huge_file,flex_bg,uninit_bg,dir_nlink,extra_isize,dir_index -i 4194304 /dev/sdc1

in order to format the /dev/sdb1, respectively /dev/sdc1 partitions.

Once that is accomplished, we can now issue the command:

blkid

which will list the UUIDs for each partition so we can reference them in /etc/fstab. For example, the output of blkid would be:

/dev/sdb1: UUID="31ced4e8-3006-4097-817e-609a4c4ca6d9" TYPE="ext4" PARTUUID="2afd8654-8d06-46da-b8f5-bed3417a6e7f"
/dev/sdc1: UUID="5cede964-0992-401f-8806-d5a3586b7136" TYPE="ext4" PARTUUID="037e0077-6c03-46c4-bd8b-3749d44c34d1"

We now create mount-points for /dev/sdb1 and /dev/sdc1:

mkdir /mnt/bag_1
mkdir /mnt/bag_2

and change /etc/fstab to mount /dev/sdb1, respectively /dev/sdc1 to /mnt/bag_1, respectively /mnt/bag_2:

UUID=31ced4e8-3006-4097-817e-609a4c4ca6d9 /mnt/bag_1	ext4	rw,noatime,nodiratime,journal_async_commit,nofail	0	0
UUID=5cede964-0992-401f-8806-d5a3586b7136 /mnt/bag_2	ext4	rw,noatime,nodiratime,journal_async_commit,nofail	0	0

and we mount them:

mount /mnt/bag_1
mount /mnt/bag_2

Now we create a mount-point where mhddfs will join the two partitions together:

mkdir /mnt/bag

and create an /etc/fstab entry:

mhddfs#/mnt/bag_1,/mnt/bag_2    /mnt/bag    fuse    allow_other    0    0

Finally, we mount mhddfs:

mount /mnt/bag

If everything went alright, then you can now write to /mnt/bag as if it were a regular filesystem and mhddfs will distribute the files to both /dev/sdb1 and /dev/sdc1.