[Linux-cluster] Sane configuration?

[Patrick Caulfield]

Nate Carlson wrote:
> Hey all,
> 
> Just getting started with the whole GFS/CLVM thing, and wanted to run a
> config by the people who actually know what they are doing before I try
> to deploy it, in case I'm doing anything really stupid.  :)
> 
> I'm basically looking at setting up a 4-node cluster, using GFS to share
> storage. I've picked up a Dell PowerVault 660F fibre array for the
> central sotrage, and a PV51F fibre switch (equivilent to a Brocade 2400,
> apparently), which will have the array and each of the boxes connected
> to it.
> 
> Here's the kicker: I'd like to run Xen on these boxes, and be able to
> migrate VM's and such between them. (This is really why I want the
> shared storage, besides just for experience.)
> 
> My plan is to create a LUN for each of the individual nodes to boot
> from, and then have a large shared LUN for CLVM. Within the CLVM PV,
> I'll create a LV for each of the Xen domains/vm's to boot from, and then
> have a large GFS LV for the shared data.
> 
> Does this sound reasonable? Anyone see any huge holes in my logic? I'm
> hoping that GFS will compile and work within the Xen-ified Linux kernel
> - I suppose I should've tried that out before ordering hardware, but ah
> well.  :)
> 
> Hope someone has some comments, and thanks for reading!

I can't comment on the particular hardware but I can say that GFS runs nicely
inside xen. There are even xenU packages for Fedora Core 4 available.

One thing to watch if you're building GFS & components from source is to make sure
that they are built with ARCH=xen on the make command-line; otherwise all
the timeouts are out by an order of magnitude!

I've done some (rather vague) instructions at
http://www.cix.co.uk/~tykepenguin/xencluster.html - there's even a primitive
xen fencing agent there, if that helps.

-- 

patrick