Re: [SSI] Run Cluster with IDE Drives

[Brian J. W. <Bri...@co...>]

"--==c.g.==--" wrote:
> 
> Hi !
> 
> Is it possible to run cluster whithout an external shared drive, e.g. with
> normal IDE drives. I know that it's not so comfortable, when adding ne nodes
> to do the whole installation on them, but for my case it would be ok.
> 
> So:
> Is it possible?
> Has anybody done it?


It is possible to run SSI clusters without a shared root, although it is
not recommended. Having a different root on each node can be
particularly problematic when migrating processes. When a process is
migrating, it has to reopen all of the files it had open on the old
node. It expects that each file should have the same pathname, the same
inode number, the same filesystem device number, etc. on the new node.
This is really only possible with a shared root.

There are some shared root solutions that would work with your hardware.
One is the GFS Network Block Device (GNBD). In this configuration, you
need one node outside the cluster with a fairly large disk. The disk
must hold not only its own root, but also a GFS root for the cluster. A
good minimum disk size is 4 GB. Sometime next week, I'll release
instructions on how to use GNBD with SSI 0.6.0. It should be a bit
easier than the instructions were for 0.5.2.

Another solution that only requires a single machine is SSI on UML. With
this solution, you would have a cluster of virtual machines. Kitrick
Sheets contributed code to 0.6.0 that makes this possible, albeit
without a shared root. It may not be too difficult to make it work with
a shared root. He and I will try to get some instructions out in the
next couple of weeks on how to do this.

A longer term solution is the Cluster File System (CFS). It's basically
NFS with tight coherency, so that the internal IDE drive of one of the
nodes can be used as the root of the entire cluster. Combined with some
volume management software, the shared root could be mirrored across the
internal drives of two or more nodes, thus providing better
availability. Dave Zafman has included the current snapshot of his CFS
work in 0.6.0. It's derived from Non-Stop Clusters (NSC) for UnixWare,
the Linux NFS implementation, and OpenGFS. It doesn't yet work, but it's
provided for reference.

-- 
Brian Watson                | "Now I don't know, but I been told it's
Linux Kernel Developer      |  hard to run with the weight of gold,
Open SSI Clustering Project |  Other hand I heard it said, it's
Compaq Computer Corp        |  just as hard with the weight of lead."
Los Angeles, CA             |     -Robert Hunter, 1970

mailto:Bri...@co...
http://opensource.compaq.com/