[Back]

G. Sheganaku:
"Optimized Auto Scaling of Elastic Processes in the Cloud using Docker Containers";
Supervisor: S. Schulte, I. Weber; Institute of Information Systems, Distributed Systems Group, 2017; final examination: 2017-12-21.

With software services becoming ever more ubiquitous, organizations are increasingly relying on interconnected business processes that act as a clockwork to orchestrate the interplay of individual services. Due to the volatility of business process landscapes, the amount of data or the number of process instances which need to be handled concurrently may vary to a large extent. Therefore, it is difficult to estimate the ever-changing demand of computational resources,
especially for highly volatile domains. With the advent of cloud computing and the virtually unlimited availability of computational resources in a utility-like fashion, organizations of any size and across industries are able to adjust their cloud-based infrastructure rapidly and on-demand. Besides applications, also complete business processes can be hosted on leased virtual resources, which enables the realization of so-called elastic processes, i.e., processes which are executed on scalable cloud infrastructure. Elastic processes can have various areas of application, e.g., manufacturing, banking, or healthcare, where flexible and scalable process support is of uttermost importance for being able to model, instantiate, execute, and monitor cross-organizational processes. Despite the manifold benefits of elastic processes, there is still a lack of solutions
supporting them. The allocation of resources for elastic processes is done on a rather high level, i.e., services are allocated to Virtual Machines (VMs) - usually in a 1-to-1 fashion, which leads to a waste of computational resources. Since elastic processes are part of potentially huge process landscapes, optimal resource allocation and process scheduling are complex tasks, usually solved by defining an optimization model. This poses a difficult computational problem, since optimization of process tasks under given constraints is a NP-hard problem.
In order to support process owners and cloud operators, this work addresses different challenges of cost-optimized elastic process enactment, by extending existing methodologies and algorithms from the fields of elastic processes, operations research, and cloud computing. This thesis presents a novel solution that enables a fine-grained allocation of services to lightweight containers instead of complete VMs, leading to a higher level of control over leased computational
resources and thereby to a better resource utilization. Optimization models and heuristics that consider resource, quality, and cost elasticity, while taking the fine-grained control into account are provided and realized as part of a newly designed container-based middleware. Extensive evaluations of the main presented approach for optimized enactments of given process landscapes under given Quality of Service (QoS) constraints show substantial cost savings and a
better resource utilization while maintaining Service Level Agreements (SLAs), when compared to recently proposed VM-based approaches.