Introduction to Linux Clustering
8.3 Five node HA cluster with DRBD + Xen
The five-node DRBD cluster design above can be extended to become a HA Xen cluster. The
three production nodes can be configured as Xen hosts, with the Xen guests being booted
from the network and using NFS for storage (just like the hosts themselves).
Cluster Suite is running on the host node, and in the event of a Xen VM failing or the entire
host itself suffering a failure, Cluster Suite can start each Xen VM on an alternate host, with
each Xen VM being a cluster node and running a single or multiple number of services.
The Xen hosts can be configured to be part of the same domain, which also allows live
migration of Xen VMs – so if one host server is being heavily loaded, some VMs can be
moved to another host whilst they are still running with no downtime at all.
Notes:
● Optionally, instead of having the Xen VMs belong to the same cluster as the hosts,
the Xen VMs could be setup in their own cluster with each Xen host running their
own cluster, leaving the host cluster to only deal with the Xen VM as a whole. See
the Xen section of this document for details on advantages/disadvantages with this
method.
Suitable Environments:
● Ideal for hosting providers
● Ideal for large companies to increase server utilisation and availability.
● Ideal for IT companies that need large number of servers for various application and
development needs.
8.4 Geographically distributed clusters
All the cluster designs show have been for use in one physical location – the cluster nodes
are all sitting in a rack, connected via ethernet, and is able to support floating IP addresses
because there is only one route into the cluster.
However, a common desire is to have geographically distributed clusters to prevent failure of
a single site taking the whole cluster offline.
Typical uses for this might be:
● A company with offices in two cities would like to have one server at each office
with the data mirrored between them.
● An ecommerce website wanting replicated email, website and database services
between two sites to ensure availability.
However, there are some big issues with a geographically separated cluster that need to be
© Copyright 2008 Jethro Carr Page 28/33
Komentarze do niniejszej Instrukcji