Partition that spans multiple disks

Question

I want to have one partition that will span multiple disks. I don't care about redundancy but in case one of the disks fails i still want to be able to access the data that is not on the damaged disk. I've read that using LVM you can span a partition over multiple disks but if one disk fails, you might lose all the data on the whole partition. RAID5 i'm not interested in it because i lose 1 disk for redundancy which is not important to me.

Note: these disks will be in a linux box.

Edit: What i want to achieve is to have a volume on which i will store various files and i want this volume to be easily expandable by adding disks to the volume. An important factor is that if i lose one disk from the volume, only the data written on that disk should be lost and not the entire volume. A 2nd factor to consider is that i don't need redundancy so a RAID5 for example is not appealing to me because i lose 1 disk for redundancy and the smallest disk is considered as base size.

Solution 1

Found what I was looking for: "[mhddfs] is a FUSE filesystem module which allows to combine several smaller filesystems into one big “virtual” one".

Solution 2

Depending on the flavor of linux you're running, you should be able to pool the disks.

• Red Hat's page on LVM pooling
• FreeBSD's page on zpool

There are also 3rd party tools that will let you do this:

• FlexRAID (commercial) will allow you to pool data and ensure all data is retained
• SnapRAID may do this as well, though I cannot confirm at this time

Comments

[terdon]

A partition cannot span multiple disks. LVM can create a logical volume across multiple disks, not a partition.

[AlexL]

Yes, it was a bad word choice.

[AlexL]

As far as i know LVM will keep the metadata spread across the disks and because of this, if a disk fails the volume might be un-usable. Is it possible with LVM to have the metadata redundant so that if a disk fails only the data on that disk will be lost?

[terdon]

Please edit your question to explain what exactly you want to do. How is having a directory on one partition and a subdirectory on another not do what you need? Why does this need to be a logical volume? Also, please remove mention of partitions because it is very misleading (and probably why you have been downvoted). Clearly explain what you are trying to achieve.

[mckenzm]

This is a very relevant question 6 years later given the availablity of capacities in the 10's of TiB's for less than USD1K. I would expect an LVM partition to span as many physical disks as it needs to (linear volume).

[AlexL]

The box is running Debian linux.

[brandonscript]

@AlexLinte which filesystem are you using?

[AlexL]

I'm not bound to a certain FS (only FS available in Debian). It can be either ext,reiser,nfs,xfs.

[brandonscript]

Then in that case I highly recommend you read over this thread: serverfault.com/questions/241659/…; it should provide you some insight into managing data over a spanned (Non-RAID) volume.

[brandonscript]

And while I can't directly give a biased answer to this, I've used FlexRAID myself and have benefited from the fact that it pools the disks on top of the existing filesystem. You may want to have a look there.

[AlexL]

I'm not looking for a commercial solution. I know the issues with LVM and that it stripes the data and metadata across the disks. If it could serialize instead of stripe that would be the solution for me.

[AlexL]

So what i want to go after is a basic disk space concatenation. If a disk fails, only data that is whole or partially stored on that disk will be lost. I'm not interested in I/O performance like striping.

[brandonscript]

Unfortunately to my knowledge there is no built in functionality with LVM to do that. You're going to need a 3rd party tool to do it -- be that open source or paid.