ScaleN - not your typical scalability model
The basic premise of infrastructure scalability is that if the component providing the scalability fails the service for which it provides scalability fails. This is not an acceptable scenario, and so high-availability (HA) architectures employing a variety of techniques were developed to avoid just such a situation.
Active-standby was the original model, in which one component (secondary) was always on standby, ready to assume duties should the active (primary) component fail. This model remains the most resilient available as it is best able to deal with the most disruptive of events, hardware failures. But for organizations valuing higher utilization over resiliency, an active-active model was adopted. This model was imperfect in its ability to ensure HA because if the total load on the combined system exceeded the capacity of a single component and the primary failed, it was assured that there would be a disruption of service but it did realize better return on investment through higher utilization rates.
Other models that addressed this limitation began to become popular. F5 adopted an N+1 best practice using N active components and a single idle standby unit. This model realizes the benefits of active-active models along with the higher resiliency of active-standby. For some time now an N+1 scalability model has been the most prevalent choice of HA architectures for those organizations desiring both high resiliency and maximum efficiency.
Traditional Scalability Models
A traditional N+1 scalability model employing a single standby and N active components.
|The basic premise of nearly all high-availability (HA) models has always been redundancy. From hardware components such as power supplies and network interfaces to network components like routers and switches to data centers, every network and application element considered critical to business continuity is duplicated.
On this model is built the traditional scalability model, N+1, which introduces additional elements and distributes load across them, with the exception of at least one standby element. The N+1 model assumes any number of active components N, with an independent dedicated secondary (standby).
The N+1 model is targeted at hardware failure and is often selected for its more efficient use of resources without radically underutilizing resources.
F5 ScaleN Model
F5 augments its already comprehensive availability solutions such as trusted N+1 HA architectures with ScaleN. ScaleN was architected with elasticity and multi-tenancy in mind. ScaleN enabled BIG-IP is as elastic as the applications it provides services for and isolates applications to enable application-level failover and scale, a capability unique to F5.
ScaleN accomplishes this by eliminating the tight-coupling between primary and standby components, allowing active devices to “fail over” to any available component configured to be part of the HA group. This includes both virtual and physical form-factors.
ScaleN also takes this capability and applies it to applications, eliminating the “all or nothing” approach to application failure associated with the N+ 1 model which deals with a single application failure by failing over the entire device – and all associated applications – to the standby device.
For many organizations managing multiple applications on a single BIG-IP this can be a disruptive event. F5 addresses this potential problem with ScaleN by allowing individual applications to failover or purposefully move between BIG-IP components. This enables resiliency at both the infrastructure layer as well as the application layer, and insulates individual applications and business services. If an application experiences a failure – whether due to hardware, software or network – it can be moved without impacting other applications.
ScaleN clustering technology also enables BIG-IP to expand its own capacity across all clustered instances to ensure its services scale seamlessly along with demand.
ScaleN supports three models for infrastructure and application scalability: Virtualization, Horizontal Clustering, and On-Demand Scaling. These models are not mutually exclusive, and can be layered and used together to implement scalability of infrastructure and applications that best fits operational and business requirements.
Stay tuned for our next post, when we'll dig into Virtualization.