Doctor of Philosophy

dissertationWe live in a big data era, which implies volume, variety, and velocity. One domain that produces and consumes big data is mobile network systems. Another domain dealing with large volumes of data is big data analytics systems known as real-time data stream frameworks. To manage big data in the network systems, the network systems should be performant to meet the real-time characteristics of the traffic. They should be flexible and evolvable to adapt to dynamics of data sources in terms of volumes in different time and distributions as well as to support new services. However, traditional network systems are rigid, which means they are not flexible and evolvable enough to adapt to changing demands or to support new services. In addition, they do not meet the performance requirements since they are designed based on basic communication and process abstractions (e.g., best-effort) without any differentiation between services and applications. In this dissertation, we first identify the limitations of the two network systems, e.g., mobile networks and real-time data processing frameworks, to provide high performance, flexibility, and evolvability. Then, we propose new designs to address the limitations in the network systems. Especially, we show how network programmability combined with multidomain information can realize performant, flexible, and evolvable network systems. The definition of multidomain information in this dissertation is a variety of data, from networks, applications, and users, which is used to inform network programmability to enhance the performance and flexibility in the network system