Modeling and simulation of data-driven applications in SDN-aware environments

Abstract

The rising popularity of Software-Defined Networking (SDN) is increasing as it promises to offer a window of opportunity and new features in terms of network performance, configuration, and management. As such, SDN is exploited by several emerging applications and environments, such as cloud computing, edge computing, IoT, and data- driven applications. Although SDN has demonstrated significant improvements in industry, still little research has explored the embracing of SDN in the area of cross-layer optimization in different SDN-aware environments. Each application and computing environment require different functionalities and Quality of Service (QoS) requirements. For example, a typical MapReduce application would require data transmission at three different times while the data transmission of stream-based applications would be unknown due to uncertainty about the number of required tasks and dependencies among stream tasks. As such, the deployment of SDN with different applications are not identical, which require different deployment strategies and algorithms to meet different QoS requirements (e.g., high bandwidth, deadline). Further, each application and environment has unique architectures, which impose different form of complexity in terms of computing, storage, and network. Due to such complexities, finding optimal solutions for SDN-aware applications and environments become very challenging. Therefore, this thesis presents multilateral research towards optimization, modeling, and simulation of cross-layer optimization of SDN-aware applications and environments. Several tools and algorithms have been proposed, implemented, and evaluated, considering various environments and applications[1–4]. The main contributions of this thesis are as follows: • Proposing and modeling a new holistic framework that simulates MapReduce ap- plications, big data management systems (BDMS), and SDN-aware networks in cloud-based environments. Theoretical and mathematical models of MapReduce in SDN-aware cloud datacenters are also proposed