Introduction to Linux Clustering
6 Combining Xen with clusters
Virtualization technology is becoming increasingly popular due to the reduced costs and better
utilisation of hardware resources.
Linux has various solutions for virtualization – one popular option is Xen which comes bundled
with a number of distributions. So the next step, is to combine the advantages of Xen, with the
advantages of clustering.
Whilst there are numerous ways this can be done (and the best solution will depend on what you
need). One basic method is to configure all the physical machines as Xen host servers and place
them into a cluster. (Ideally, the host servers should also be in the same Xen domain to allow live
migration of VMs from one server to another.)
The cluster management software can then be configured to treat virtual machines as services –
moving them between cluster nodes and restarting dead VMs on other machines.
However, this only provides basic failover services – a VM will only be moved to another host if the
whole VM becomes unavailable, or if the host node crashes.
For more fine-grained control, there are two options:
6.1 VMs as part of main cluster
Add all the virtual machines to the cluster as nodes, along side the physical servers. The
services can be configured using failover domains to only failover to virtual machines.
This means you only have one cluster, but it will introduce a lot more complexity into the
cluster configuration and administration.
6.2 VMs runs as a separate cluster.
The other option is to run the cluster software on the Xen VMs themselves. This can be
useful in that it allows you to configure multiple clusters on top of the main cluster which
may be appealing to hosting providers who can offer customer their own private two-node
clusters.
This method is also useful for systems that intend to run large number of services on top of a
single VM and provides the ability to migrate individual services from one VM to another.
The disadvantage is that more overhead is introduced by running the additional clustering
software on all the nodes and it may become more time consuming to manage.
© Copyright 2008 Jethro Carr Page 15/33
Komentáře k této Příručce