Motivation

Many times during my professional career, I found that being able to use a bunch of servers on a dedicated network is very convenient to perform tests and benchmarks of software for infrastructure provisioning, management and monitoring. When starting to work at CETIC, I had to set up and operate inGRID (in French), a cluster dedicated to Reearch on Grid Computing. I also had the opportunity to design a second cluster (in French) targeted as Research test bed to study, test and contribute to development of Distributed Architecture and Cloud Management software.

These experiences confirmed my idea of ​​setting up my own cluster to test new technologies and concepts like alternatives to Docker for containerization and software from the “Kubernetes galaxy” for containers orchestration, scalability and redundancy. Such cluster is also a training ground for more widespread technologies in the field of Linux system administration like automated infrastructure management and system monitoring.

There should be enough nodes in the cluster to test concepts like high availability or load balancing. Three nodes seems a minimum and six nodes should allow partitioning to simulate inter-clusters communication. Therefore, such “minicluster” should contain at least six machines.

Technical information is available in a repository.

The story, so far

• A first idea of the logical architecture of the cluster.
• Choosing cost savvy hardware: it is an experiment cluster.
• Provisioning of all nodes of the cluster with a Linux distribution in an unattended and automated manner.