Part II: Overview
Overview of the Diskless Shared Root Cluster
PART II: OVERVIEW
Today's IT infrastructures have to be optimized for cost efficiency, high availability and performance. Reaching these goals means great advantage over competiting companies.
The Diskless Shared-root Cluster provides an enterprise IT infrastructure which optimally complies with these requirements.
A Linux cluster filesystem is the core of this solution. It combines Linux server and attached storage networks to form a Single System Image (SSI) on filesystem level.
By seperating storage from the server layer the whole IT infrastructure is granular scaleable. The system can be scaled by independently adding storage capacity respectively CPU/IO performance in the storage or server layer (2D Scalability).
In the past such sophisticated technology was only available for UNIX as proprietary products. With the Open-sharedroot project it is now released as Open Source. As with most GPLed software that is used in huge enterprise scenarios, professional support services are available as well.
The platform usually consists of two or more servers (the nodes) that contribute their work to perform different tasks like storage clusters, high availability solution, application loadbalancing or high performance.
Storage clusters provide access to a shared file system (e.g. Global Filesystem GFS). This removes the need to replicate data between different storage servers. Clients may be distributed among all cluster nodes to loadbalance their connections. Both read and write access is possible. The cluster takes care of filelocking to circumvent any data corruption. In this scenario a Diskless Shared Root Cluster greatly simplifies administration, backup and disaster recovery because all nodes not only share the same data but also share the same configuration.
High availability clusters are used for mission critical services that must run continuously. It is mandatory that such clusters elimitate all single points of failure and provide a consistient configuration on all cluster nodes so that a service may relocate if a node fails (a.k.a. failover). A Diskless Shared Root Cluster share the same configuration and data on all nodes. Therefor this configuration is very suitable for failover clusters.
Loadbalancing clusters distribute client connections among all cluster nodes. This greatly increases scalability because the nodes are faced with an even load. As a speciality a Diskless Shared Root Cluster may run a loadbalancer cluster service. Thus it is not necessary to buy special loadbalancing hardware.
High performance clusters (HPC)are used as number cruncher for parallel calculations. While some applications handle the distributed calculation within user space, usually you have to enable HPC with kernel tweaks. As the Diskless Shared Root Cluster is built for enterprise environments we stated the rule that there are no binary modifications for system files possible. While this is mandatory to maintain a valid support entitlement this also prevents some HPC methods. However as a Diskless Shared Root Cluster in many respectives behaves just like a regular RHEL server you most like may enable HPC - alas you may loose support.
If you construct your cluster it is of course possible to mix the cluster types.