An evaluation testbed for adaptive, topology-aware deployment of elastic applications

Abstract

Cloud application providers who deploy their application at different cloud sites usually aim for close-by processing of user requests, benefiting from improved quality of service, and traffic reduction [4]. In this context, we dynamically scale applications to reduce costs by automating their deployment and adapting their resource allocation dynamically. We research the following questions: Where to allocate how many resources and how to apply the allocation? Which information is needed and how to exchange it? How can applications cope with ever changing resource allocations? To practically evaluate our solutions, we created a flexible testbed. We share our insides and implementation to researchers tackling the diverse subproblems, various optimization goals, potentials for cost savings, and QoS improvements. We provide software with install instructions to construct your own private testbed [5]. Our testbed is two-layered: The bottom layer allows to test VM deployment on emulated, geographically distributed sites. It can be independently reused, is self-sufficient, and thus constitutes a small testbed on its own, the GeoDist Testbed (Section 2). The top layer allows to test adaptations and VM placement algorithms interactively or through predefined scenarios. Both layers together form the Adaptation Testbed (Section 3). Its capabilities are demonstrated in three different scenarios (Section 4).