The trend for most businesses is to provide their services or sell their products online. By tapping the internet, they would be able to expand their customer base not only in their physical location but globally as well. Other businesses that can’t provide their services or sell their products online still use the power of the internet to provide additional information to their prospective customers.
The data, products and services they provide have to be available to their customers 24/7. A downtime for businesses means loss of precious customers and destruction of company image. A website or an online store is basically the face of the business to its customers and unreliability will easily translate to the company’s image.
But maintaining these services to prevent any downtime is difficult when only a single computer handles the requests and processes. For that reason a number of computers have to be clustered to be part of larger component that will assure availability no matter what. Aside from simply connecting them together, this form of clustering assures availability of the functions even if one of the nodes fails to work.
This form of clustering is referred to as High Availability Clustering. Each component or node has the ability to provide the needed function so that it can easily take over when one of the nodes fails. The nodes are often referred to as “redundant” since each of them mirrors each node’s functionality. HA is also called as “failover clustering” as the design assures stability.
Forms of HA Clustering
HA Clustering could be differentiated based on the behavior of the nodes.
The most common form of HA clustering is the active/active configuration. In this type of HA clustering the nodes are programmed to share the workload. However, when one node fails to function, the workload of the failed node will be distributed to other nodes until the failed node is fixed.
Next is the active/passive configuration. The nodes in this configuration all have redundant functions wherein they will take over the primary node in case any problem arises. Basically, only a few nodes will be functioning while others are just back-ups.
The N+1 is considered an active/passive configuration – with a twist. The active nodes share the same workload while there is one node that has the capability of doing everything. However, this type of node is not actively used and will only be tapped to replace the node that failed to work.
N+M could be considered as the more expensive version of N+1. Instead of another node to replace the failed node, a standby server is utilized. M refers to “more” servers which will help in node balancing.
When HA clustering uses N to 1 node configuration, the “extra” node is configured to become universally adaptable but only to a certain extent. N + 1 replaces the failed node while N to 1 is only a temporary configuration.
The N to N configuration is almost the same as active/active configuration. However, an extra node is provided for every failed node to return to the original configuration without going into standby mode.
Although there are different forms of HA configurations, there are general requirements in HA to ensure its success.
• Ease of Start/Stop Operation – this could be manual or through a program but what’s important is that there should be an operation that will immediately stop a node if a problem is noticed.
• Availability of Storage Area Network – this is to ensure data availability for all nodes.
• Ability to Save State – the HA configuration should have the ability to continuously save the state of cluster so that the latest state could be implemented when a node is implemented. This also ensures that the entire network will keep the data even if it crashes.
More than Just Data/Function
High Availability clustering is not just about shared data and functions. Although the main function of HA clustering is to ensure the data is available for the users and function continues to work, hardware configurations should also be shared. Cable connections and power sharing should also be implemented. Without these physical connections, HA clustering will never function or fail to ensure consistent efficiency as more nodes will fail to work.