Network Resource Optimization in Cloud Computing Environments
Konstantinos Oikonomou (supervisor)
To address the problem of managing particularly dynamic environments, such as cloud computing environments, where resources and services are created and destroyed dynamically based on certain parameters, such as the amount of data at any given time in the network, users’ needs, or the energy consumption, new approaches are needed, which take into account those parameters in order to provide the appropriate infrastructure, where network functions and services are deployed and executed as applications in suitably customized virtual machines. In such environments special emphasis is placed on the selection of the optimized location for the corresponding installation, as well as on the applied communication and routing protocols.
Based on the above, present PhD dissertation focuses on procuring answers to research questions regarding the optimal location placement of virtual machines in such environments, with the aim of minimizing network costs, energy costs, maintenance costs, etc., whilst enhancing the overall user experience and quality of service. Consequently, the dissertation’s main objective refers to the study and analysis of the specific problem and the formulation of automated distributed approaches for placing, consolidating and migrating virtual machines from one network node to another according to the network’s current state.
Beyond these, important field of research in current dissertation are the opportunities arisen from new technologies, and especially those offered by smart mobile devices. These devices, due to their increasing computational capacity and capabilities, may potentially play the role of Micro-Data-Centers, integrating cloud networking services at the edges of the cloud, and therefore, adopting characteristics of the fog computing environments, leading to a significant increase in the system’s scalability and total complexity.
The aforementioned research challenges are considered as facility location problems, which by nature are of non-polynomial time (NP-complete), and require local network information. Although, there exist approximate algorithms for solving them, the majority of these approaches make use of centralized methods with requirements for global knowledge of the system’s parameters and topology, making their use in many cases prohibitive and unsuitable for large-scale environments. It is therefore necessary to consider distributed approaches, based on local information, capable of automating network processes and offering system flexibility and adaptability.
In conclusion, current PhD dissertation addresses the problem of discovering the optimal facility locations for housing virtual machines, which embed various services, in cloud and fog computing environments, with the goal of minimizing the overall network costs. This problem cannot be totally resolved by traditional centralized algorithms that demand global knowledge of the network’s topology and variables, since in cases of system changes or modifications, they need continuous recalculations in order to effectively respond to user demands. In this context, traditional centralized approaches are not considered appropriate for large-scale dynamic environments such as cloud computing networks, in terms of system complexity and global information requirement. In present PhD dissertation, the problem of determining the optimal location for placing a virtual machine is being reviewed with methods that are both scalable and dynamic, since their focus is on distributed approaches, using local information to decide the best suitable location for installing a service. The main purpose is to create analytical models that efficiently describe and address the problem of optimal virtual machine placement in environments that correspond to real conditions. The analytical models will offer a qualitative analysis of the above-mentioned problem, while experimental simulations will provide an evaluation of their effectiveness. Finally, experiments under real conditions (testbed demo), using as a case study modern smart applications found in smart cities and virtual reality environments, will also take place and produce a comparative assessment in correlation with the analytical models and the generated simulation results.