The Art of Capacity Planning, 2nd Edition

CHAPTER SIX

Autoscaling

In Chapter 5, we talked about installing hardware in your on-premises datacenter and its deployment to production. As it is well known, use of public cloud obviates the need for this. Further, it obviates the need for maintenance of the infrastructure and thereby boosts product development agility, which is crucial in an increasingly competitive landscape. For instance, in 2008, Netflix kicked off its migration from on-premises to Amazon Web Services (AWS), and in February 2016, it announced the completion of the migration of its streaming service to the cloud.¹ Migration to the cloud eliminated the cycles spent on hardware procurement and datacenter maintenance, and resulted in higher development agility.

The use of a cloud service at large scale is much more expensive compared to the use of an in-house datacenter. This calls for the development of techniques to minimize the cost overhead associated with the use of a public cloud without sacrificing its various benefits such as elasticity. Autoscaling allows automatic scaling up of capacity when it is needed and scaling back down when it is not needed. An enterprise can set a minimum required capacity across Availability Zones (AZs)—isolated locations in a given geographic region—to ensure quick accessibility. This helps save cost by providing the best availability and performance for a given cost. For cases in which a reserved capacity is purchased, the cost can be amortized by scheduling jobs of noncritical ...