Increasing performance and resilience by clustering load balancers
The variable nature of many application workloads requires an application delivery infrastructure that can easily scale to meet the changing capacity demands. When planning to load balance a workload that has variable, or possibly unknown, future capacity requirements, IT architects need to ensure that the application delivery infrastructure can adjust to meet these requirements. Load Balancing Clustering, introduced with our January 2016 release, provides IT architects with a solution that can easily scale to meet future requirements by simply configuring multiple appliances to act in parallel to deliver higher application delivery capacity. Clustering also brings high availability as a bonus – if a cluster node fails, service will continue to be provided by the remaining cluster nodes which significantly enhances application uptime metrics.
Without clustering, capacity planning can prove difficult as an over-specified solution is a waste of resources while a solution that cannot accommodate growth will require a costly replacement as demand exceeds capacity. With clustering, the capacity planning process is simplified as there is a clearly understood growth path that can be costed accurately and allows the right sized capacity to be deployed at the right time.
Keeping it Simple
Up to eight LoadMaster appliances (including bare metal and virtualized) may be grouped together in a cluster to form an entity which logically appears as a single LoadMaster instance to clients, servers and administrators. Because the cluster appears as a single instance, there are no special infrastructure requirements (such as configuring upstream router to support ECMP) and administration is exactly as on a single appliance LoadMaster. Taking this low-impact approach to clustering makes migration from a single LoadMaster to a cluster very easy and removes complexity from the networking infrastructure.
Clustering and High Availability – What’s the difference?
In the KEMP LoadMaster world, High Availability (HA) is used to describe the scenario where two LoadMaster appliances are deployed in an active-passive configuration (a HA Pair) where the passive appliance will become active if the currently active appliance fails. This approach provides resilience against failure of a single appliance but as one appliance in the HA-Pair is idle, the maximum throughput is the capacity of a single LoadMaster. In contrast, in a cluster all appliances are actively processing traffic which means the throughput is an aggregate of the capacity of member appliances. A cluster will also provide resilience in the case of a member failure as remaining nodes will continue to provide service. So which approach is better? There is no simple answer to this question as each deployment scenario will have differing requirements for growth and sustained capacity in the event of failure. In general, HA is appropriate for environments that require resilience and preservation of capacity while clustering offers resilience plus the capability to accommodate very large workloads and to grow to meet application demands.