[Linux-cluster] Multiple Logical Volumes

[Robert Peterson]

isplist at logicore.net wrote:
> Although, I'm now looking at the diskless booting ideas which might make a lot 
> more sense for my needs. From what I can tell, I can have multiple roots even 
> with a single combined storage system? I'm not sure, still reading, need to 
> try some basic two node stuff maybe.
>
> Thanks very much, another one for my notes.
>
> Mike
>   
Hi Mike,

Well, I have some experience with this as well, although that was in my 
pre-GFS days.
I was using nfs to serve multiple root directories.  So here's what I'd do:

1. Set up your dhcp server to dole out unique IP addresses to your 
clients based on
MAC address.
2. Mount your GFS partition as /tftpboot
3. In your /tftpboot, create root directories for your clients, and put 
what you need there.
In my case, I had tiny root partitions and my diskless clients were tiny 
(literally) embedded
devices that had to run something like Denx's ELDK, The Embedded Linux 
Development Kit,
(google it) which included tools to get their kernel from tftpboot, and 
boot from the tftp server.
My client root file systems also had separate mount points for commonly 
shared areas that
didn't tend to change a lot: /lib, /usr/lib, /bin, /sbin, and several 
more, etc., so that if I changed
one, I changed them all.

You still need a minimal set of these libs and such on the client root 
fs in order to boot far enough to
where the fstab can actually mount these things though.

Also, the clients need to have enough smarts to each mount their OWN 
root partition
so they don't bump into one another.
4. Of course you need nfs serving out the data, and xinetd accepting 
tftp connections.

For extra credit, you could use a NFS failover service through rgmanager 
for your NFS server.  :)
I wrote a little bit about this in my NFS/GFS Cookbook, but not in much 
detail.

I hope this helps.

Regards,

Bob Peterson
Red Hat Cluster Suite