2

a short example of how the infrastructure may look like:

4 servers act as iscsi targets and expose 1x2TB each (every of those servers has a raid10 for the storage, thus 2TB of usable capacity)

n servers that need to have an 8 TB (consisting of the exposed disks above: 4x2TB) disk attached to them, concurrent writes are necessary. All servers really share the storage.

Now, how to achieve that? I do not have any preference regarding the clustered fs.

regards, Patrick

3 Answers 3

1

DRBD spread across two servers both acting as iSCSI targets to the same logical block device works pretty well and is relatively simple. Simple, cheap solutions like these are where DRBD shines and it's been included in the stock Linux kernel source for awhile now.

http://www.drbd.org/

0

I implemented something similar for this need just yesterday. Our situation was similar. 3 machines total, 1 was the 'head' and 2 were simply serving up their data via iSCSI target. I used openfiler as the head. They do not officially support mounting iSCSI and re-serving them as native NFS/CIFS however it was trivial to log in via ssh, mount them, and proceed normally within the openfiler GUI to add them to the volume. If you use a versatile file system (XFS), combining them all and expanding over time should be pretty easy.

Under the hood, this is nothing more than openiscsi+lvm+samba/nfs but it tends to work pretty well.

I did, however, have some issues since a) its not supported and b) its a bit kludgy. I was able to get everything to start properly on reboot, including mounting the downstream iSCSI targets and having it serve properly from a reboot, however I expect complications should an individual SAN become unavailable to the rest of the kit.

Good luck

1
  • I agree that this is possible, but having very high availability requirements in conjunction with 'unsupported configuration' is doubleplusungood. Feb 6, 2011 at 21:52
0

I would recommend using OCFS for your purposes. What I do not understand in your configuration: What do you do if one server fails? Where is the redundancy for that?

You must log in to answer this question.