Clustering is a way to provide hardware resilience, by removing single points of failure. As far as possible the machines used should use resilient internal components, mirrored disks, duplicated memory, dual networks, power supplies etc. They are then connected together by a pair of heartbeat cables (another network). The dual networks connections and the 2 heartbeat links need to be split across multiple NICs again to avoid single points of failure.
All the storage for whatever applications they are running is held externally in a pair of disk arrays or a SAN, this allows either machine to act as master node and control the data. Should that machine experience some major problem, the other machine takes control of the cluster and the data.
There are 2 main configurations active/active and active/passive.
In A/A each machine runs some of the services normally, but takes everything if required. Clearly the machines must be capable of running the whole load.
In A/P mode, one machine runs the load and the other is idle until required.
More advanced clusters use more than 2 servers or are split across datacentres.
Clustering and resilience generally is a trade between cost and risk.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.