Self-Adaptive Resource Scalability for Elastic Service Provisioning in Cloud Architectures Pooyan Jamshidi, Aakash Ahmad, Claus Pahl Lero – the Irish Software Engineering Research Centre School of Computing, Dublin City University, Ireland. [pooyan.jamshidi *, ahmad.aakash, claus.pahl]@computing.dcu.ie Research Motivation Cloud computing exemplifies a computing paradigm, where resources are provisioned as a service (Armbrust, et al., 2010) (Jamshidi, et al., 2013). A key distinguishing factor in cloud computing, other than in similar paradigms like Grid or High Performance Computing, is the provision of service-level agreements (SLAs) to service users (Brandic, 2009). Thereby, applications can function considering predefined non-functional requirements such as execution time, cost, security or privacy standards. However, due to changing services, workload, external environment, hardware, and software failures, established SLAs may be violated. On the other hand, application providers (i.e. consumers of cloud-based services) want to minimize their cost while meeting the SLAs (Ghanbari, et al., 2012). As a result, application providers must interact with their systems to adjust their cloud resources. However, this frequent user interaction lead to a barrier for successful service provisioning of cloud-based infrastructures. We identify the central research challenge as “how to autonomically scale an application in order to satisfy the non- functional requirements of software systems while minimizing cost of service provisioning through multiple cloud providers and considering the involved uncertainties?” State-of-the-art An elasticity policy (Vaquero, et al., 2011) governs how and when resources are added or removed from a cloud-enabled software system. Automated cloud application scalability (called auto-scaling) is one of the most recent development for dynamic and automated resource provisioning (Vaquero, et al., 2011) (Ghanbari, et al., 2012) (Shen, et al., 2011). Auto-scaling is responsible for implementing such application’s elasticity policies. The most recent auto-scaling capability, which is available for the commercial cloud providers (e.g., Amazon EC2 or Windows Azure) enable rule-based specification of elasticity policies (Vaquero, et al., 2011) (Ghanbari, et al., 2012). However, the default auto-scaling in the cloud providers may not necessarily optimize the cost incurred from the allocation of resources to the application (Ghanbari, et al., 2012).