We help IT Professionals succeed at work.

Moving NFS servers to fibre-SAN.

NigelW1 asked

I have inherited the following setup.

2 Linux servers running NFS in an active / standby configuration with heartbeat software
to manage the failover.

DrDB mirroring the data from one server to the other.

I would like to move to using a fibre-SAN (already available for other reasons).

So my questions are.

1) What is the recommended storage configuration – ‘shared storage’ between the servers
     -  no DrDB ?

2) How do you manage the file system across the 2 servers ?
    Do you need to worry about mounting / dismounting the file system on the active / standby
    nodes ?

Thanks for any help.

Note: I am a novice when it comes to NFS / SAN technology.
Watch Question


My thought is this - right now you use DrDB just to keep the 2 sync'd in case one goes down, but not really for redundancy.

Once you get into the SAN arena, all the redundancy/fault tolerance is built into the arrays.  Now, a lot of SAN solutions are able to support NFS natively, so if that describes your circumstance, you could ultimately do away with the Linux machines entirely.

But I'm thinking you can actually re-use quite a bit of what you have now, with some caveats.

See if you connect each Linux server by fiber-channel into your SAN into a single LUN on the SAN, the first won't be able to see changes made by the second, and vice-versa (unless you put some kind of global filesystem on there with a 3rd party mediator server - I don't recommend it.)

However, if you go fiber into the SAN from each box, and each box has its own dedicated slice of SAN, you can continue to run the NFS servers in active/standby, and then use DrDB instead to mirror the two SAN LANs rather than the Linux servers' local drives.

You get to waste no hardware but get out of the internal spindle business, and minimize your single points of failure.  What do you think?



Thanks - this sounds like keeping two copies of the data on the SAN - which may well be the easiest way to go from where I am.

A couple of questions.

1) "but not really for redundancy" - can you explain a little more.

2) "global filesystem on there with a 3rd party mediator server" - I know you did not recommend this option but for interest can you give an example.
1)  I guess the more I think about it, your DrDB use really _does_  provide redundancy - my original comment was that it struck me as more of a sync operation than true RAID or non-software level mirroring which I'd classify as redundancy.

2)  An example is GFS.  You would basically configure 1 LUN on the SAN which both the former NFS systems would be fiber-connected to.  Each fiber channel client now mounts the GFS system as a local drive, but every time any writes, it sends a signal to a mediator daemon that you must set up (can be on a 3rd server, or just on one of the 2 existing units)  In this scenario you introduce another point of failure, and the performance is questionable on writes.  I think its somewhat of a hack way of sharing a LUN, and I'd personally just mirror a couple of SAN slices.

Explore More ContentExplore courses, solutions, and other research materials related to this topic.