Software-Defined Networking 2.0

日益增长的网络需求要求当前的互联网发生重大变化。软件定义网络（SDN）的提出就是为了应对这些挑战。SDN背后的基本思想是将控制平面与交换机（数据路径）分开。这种分离本质上可以：(i)使网络交换机极其简单、快速和廉价；（ii）由于更改是在软件中完成的，因此易于对网络控制器进行更改；（iii）通过创建网络资源的全局视图来简化网络管理和优化；（iv）降低个人和小公司建立和改变用于大规模网络的网络组件的进入门槛，这是降低成本的另一种方式。

The ever-increasing demand for networks requires significant changes in the current Internet. Software-Defined Networking (SDN) was proposed to address these challenges. The basic idea behind SDN is to separate the control plane from the switches (data path). This separation can essentially: (i) make network switches extremely simple, fast, and inexpensive; (ii) make it easy to make changes to the network controller since the change is done in software; (iii) simplify network management and optimization by creating a global view of network resources; and (iv) lower the barrier to entry for individuals and small companies to build and change network components to be used in large-scale networks, which is another way to reduce cost. A major accelerator for development and deployment of software-defined networks has been the OpenFlow protocol. This is the de facto standard for interactions between network switches and SDN controllers. Unfortunately, OpenFlow intrinsically limits the information channel between switches and the control plane, which imposes significant restrictions on the types of controller applications we can support. For example, any applications (e.g. smart firewall, application-aware load balancer, etc.) that require access to payload of packets beyond the initial packet of a flow cannot be implemented with OpenFlow. Can we move from a rigid flow-based protocol like OpenFlow, to a design where control applications can define and dynamically change the granularity and the pattern of packets they receive from switching elements? There are two new advances in SDN world that might make this possible: 1) Recent developments in programmability of network devices (e.g. P4) allow execution of a complex logic on every packet in data plane. This is a very powerful feature that gives extreme control on which packets to send to the controller; and 2) with the introduction of the new generation of distributed controllers (e.g. ONOS, and Beehive), scalability is not considered to be a major roadblock for deployment of SDN solutions. These advances open the door to new protocols that allow control applications to tune the granularity of the data they require from switches, which we call "Software-Defined Networking 2.0". SDN 2.0 will allow a wide range of new applications that cannot be implemented in today's networks. A smart firewall application, for example, can start with a low overhead surveillance of the entire network, and dynamically increase the number of packets sent to the controller when it identifies abnormal activities in certain parts of the network. SDN 2.0 can be used for debugging and troubleshooting network control applications, a feature missing in today's networks. To that end, we can increase the fraction of traffic sent to the controller for fine-grained analysis and troubleshooting, and reduce the load under normal operation of the network.