| Abstract |
A fundamental challenge of programming an SDC control plane is how to handle both operating environment (e.g., topology) and user-control-programs (e.g., access control policy) dynamics automatically. High-levels of dynamicity, limited resources, and the need to achieve unified control of both networking (i.e., SDN) and computing (i.e., network functions) resources, make the task extremely complex. In this paper, we conduct the first systematic study of unified, automatic SDC pro- gramming. We first show that integration of asynchronous and continuously changing SDCs introduces substantial complexities. We then present Magellan - a novel, unified, automatic SDC programming framework. In particular, it first introduces a set of high-level programming abstractions including stream attributes, live variables, and route algebra, and thus allows an SDC programmer to express SDC control objectives. It then presents update algebra to execute SDC control programs efficiently, and update automatically, despite resource constraints such as weakly connected control-data planes. We conduct extensive evaluations of Magellan to demonstrate its efficiency. Preliminary results appeared in SIGCOMM 2018 and INFOCOM 2019. |