Hi,

Just curious... Is't possible to setup a one(1) tier cluster that has load balancing and high availability using Redhat cluster manager suite (luci/ricci)? I understand that this suite sets up high availability clusters only.

While single node clusters with different cluster services are possible with Red Hat Cluster Suite and we make heavy use of it for development systems, the conga interface (luci+ricci) requires imho at least two nodes.

I concur. The reasons we installed it in the past were:

1- We wanted redundancy in our web application layer
2- We wanted a strong file lock mechanism where the same files were available to the 2 machines

Why pay for the cluster suite if you have only one machine ? How can you offer HA with one machine ?

The single node clusters are used mostly for quality assurance chains like crash&burn, development, quality assurance, pre-production, production that are required by certain software solutions (e.g. SAP, Webservices have usually less stages).

In our case the cluster nodes concurrently use a shared root filesystem that is formatted with a cluster filesystem called GFS. Thus all nodes are able to use the same root filesystem with read/write access. This requires an installed cluster suite.

At first this may seem costly but there are big advantages in the management, deployment, backup and desaster recovery of clusters.

For example we handle different quality assurance stages automatically by simply cloning the root filesystem together with the cluster meta-data and create single node or regular clusters of any size by booting nodes from the same or from cloned root filesystems.

If the tests are successful, we clone them to the next stages until they are released as pre-production or production ready cluster. During the cloning process we can again decide if we clone to a single node cluster or if we want to deploy the cluster on more nodes.

Single node clusters have no HA capabilities but with the cloning we can guarantee that all machines have the exact configuration and we may roll out the settings to any cluster sizes.

There is also the possibility to use CentOS for the less critical clusters and use kickstart to create productive RHEL clusters from the CentOS configuration.

The cluster nodes have no internal harddisks and are used as general purpose nodes. By switching the boot device they may instantly join different clusters and thus can be used in a very flexible way.