I'm trying to figure out if this is a stupid idea or not, comments please:
I have a small, somewhat aging, 4-node cluster (legacy of a former
web-hosting business), that currently hosts a few web sites and a
mailing list server for some local community groups. I've also been
using it as a development sandbox, and playing with technology a bit.
Right now, I have it set up for poor-man's high availability:
- each node has 4 2TB drives, and 2 gigE NICs - 1 net is used for
outside communication, 1 for node-to-node stuff
- xen virtualization, with a couple of VMs for production, and others
for development/playing around
- md RAID-10 on each node
- DRBD across pairs of nodes to support the production VMs, with
pacemaker-based failover
I keep looking for ways to get beyond the pairwise nature of DRBD, but
don't quite have the horsepower or memory to spare to run something like
Gluster to provide replication across a cluster file system.
The following thought keeps coming to mind: what about mounting drives
using AoE, then running md RAID across the nodes?
- use LVM to create and manage logical volumes on each node
- publish a group of logical volumes via AoE (e.g., 1 from each of the 4
servers)
- use md within a VM, to assemble a RAID10 array
- if a node fails, taking down the VM, it's instance of md, and the
volume on that node, failover to another node, bring up the same VM, and
it's instance of md will re-assemble the array from the 3 remaining nodes
Does this make any sense? Will it work with any kind of performance at
all? Am I all wet?
Thanks for any comments.
Miles Fidelman
--
In theory, there is no difference between theory and practice.
In practice, there is. .... Yogi Berra
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
