Towards practical auto scaling of user facing applications

Cloud Computing has the purpose of providing computing services in different levels, from remote data storage and computing resources (Infrastructure as a Service - IaaS) to applications accessed through the Internet (Software as a Service - SaaS). From the perspective of an application provider, it is important to manage the capacity of the applications being offered. Typically, such applications are long-lived Web-based applications that present highly variable workload, which is difficult to be predicted accurately. In order to manage the capacity of such applications efficiently, application providers have two options: run the applications on a statically over-provisioned infrastructure that is able to handle the expected peak load of the applications, or acquiring resources on an on-demand basis from IaaS providers. In this paper we pursue the later option. We aim at investigating the completeness and usability of a new service offered by IaaS providers, which has being known as "auto-scaling". This service allows the configuration of capacity management policies that must be applied to dynamically decide on acquiring or releasing resource instances for a given application. Policies like that have been studied in the last decade by researchers in academia. We here try to shed some light on the plausibility of using the new auto-scaling service to implement policies defined by researchers. To this end, we evaluate an implementation of important dynamic provisioning policies onto the auto-scaling service and the cost of such service, trying to finally find out a link between the current cloud market and the studies on dynamic provisioning of resources being carried out in academia.