Currently, software-defined networking (SDN) has become a fashionable topic in the industry. SDN technology and its possible impact are highly valued by academia and industry, and there are also different understandings and understandings of SDN. This article attempts to comb the development context of SDN technology, analyze the connotation, essential characteristics, application fields, and development trends of SDN technology, and then analyze the impact of SDN technology development.
1. "Different opinions" SDN
At present, the experts who pay a lot of attention to SDN can be roughly divided into three categories: IDC design and operation and maintenance personnel, data equipment design and development personnel, and future network research and test personnel. These three types of experts pay attention to SDN with different starting points and different understandings of SDN, and their development visions and expectations for SDN are also different. After the emergence of a new technology, various discussions on tracing back who put forward the concept, when it was put forward, and whether it is a new concept are usually of academic significance only, and do not have any substantial help in understanding the nature of the new technology. Discussing the concept of SDN still has to start with its real market demand.
(1) Commercial demand for SDN first appeared in the data center
In order to support the migration of virtual machines on the application server, IDC's internal network is usually a two-layer network, because if a three-layer network is adopted, then when the virtual machine is migrated, the IP address of the application service corresponding to the virtual machine will change accordingly, bringing business Difficulty in deployment and management, but the second-tier network does not have this problem.
However, the direct application of the existing Layer 2 network technology within IDC will mainly bring about the following two problems: One is that in the Layer 2 network, in order to eliminate the loop of broadcast packets, the Spanning Tree (STP) protocol is usually used. , A logical tree is constructed between network nodes, and the traffic between nodes is transmitted according to this "tree-like" topology. Even if there are multiple physical links between network nodes, only one link actually transmits data, and the other links The roads are free (only for backup). However, within IDC, there are frequent data exchange requirements between multiple servers. STP-based tree network topology cannot efficiently support this "horizontal" traffic. The idle links between servers also cause a large amount of network resources. Waste, so the STP-based Layer 2 network is too simple for IDC and needs to be reformed, especially with the development of cloud computing, the need for such IDC internal Layer 2 networking is becoming more and more urgent. Another problem is that there are usually many application servers in IDC, some of them reach tens of thousands or even hundreds of thousands. The second layer switch needs to use ARP and other protocols to learn the source address of the received data packet to build the MAC address table. There are many application servers, so there are many MAC address table entries, which usually exceed the capacity of the conventional two-layer switch MAC address table, which will cause a large number of MAC addresses to be unable to enter the MAC address table. The two-layer switch is unable to find the MAC address in the MAC address table. The data frame corresponding to the address is broadcast in the Layer 2 domain, causing flooding of traffic within the Layer 2 network and affecting the efficiency of the IDC internal network.
The root cause of the above two problems is that the traditional two-layer network design is too simple, the two-layer switch will only learn the MAC address, and will not plan the data forwarding path based on the MAC address, that is to say, there is no traditional two-layer network. The control plane (or the function of the control plane is very, very weak, and is integrated with the forwarding function), and only the data plane (responsible for the forwarding of data frames). Therefore, it has become an urgent need to increase the control plane (or strengthen the control plane function) in the second-tier network, and be responsible for the traffic scheduling and management between the internal nodes of the larger second-tier network. The current mainstream solution is to use variants of the IS-IS routing protocol to construct the control plane routing function; use Openflow to define the interface between the control plane and the forwarding plane. This leads to the concept of separation of the control plane and the forwarding plane. But this is only the separation of the control plane and the forwarding plane in the Layer 2 network.
(2) SDN commercial demand comes from the optimization of router internal functions
In traditional routers, the interface between the control plane responsible for routing planning and routing strategy and the data plane responsible for data encapsulation and high-speed forwarding is not open and is tightly coupled. Each manufacturer connects the control plane and the forwarding plane through its own protocol or interface. This is also where advantageous manufacturers such as CISCO and JUNIPER maintain technical barriers and exclude emerging manufacturers.
However, there are two forces that are quietly challenging this model: one is large Internet companies, which have the need to build their own corporate networks, and these Internet companies believe that the current communication needs of their own corporate networks are specific, while the traditional The functions of routers are too complicated. More than 80% of the functions and features are not available in your own network. However, when you buy these routers, you have to pay for these useless functions. It feels "innocent", so there is self-designed and simple implementation. The need for efficient routers is also one of the original intentions of companies such as Facebook, Google, and Yahoo to establish ONF (Open Network Forum) to develop the SDN standard. Because these Internet companies have a large number of successful experience in using their own customized application servers within IDC, they have good expectations for independent research and development of efficient routers. Another force is emerging data equipment manufacturers. They try to break the tight coupling between the control plane and the data plane inside the router to form an open and standard device interface, so that the control functions can be centralized and separated separately, so that the data The forwarding equipment can be made more versatile and simple, and the cost can be made lower, which will help break the monopoly of CISCO, JUNIPER and other manufacturers, and these manufacturers can obtain new development opportunities from it.
Based on this consideration, the IETF carried out research work on the separation of the internal control plane and the forwarding plane of the router earlier, and established the FORCES working group to define the communication protocol between the internal control plane and the forwarding plane of the router. Although this also leads to the concept of separation of the control plane and the forwarding plane, this is the separation of the control plane and the forwarding plane in a three-layer network.
(3) SDN commercial demand comes from future network research and testing
At present, in order to solve the problems of insufficient network address space, difficulty in guaranteeing service quality, lack of security and credible mechanisms, and poor network management and control capabilities faced by IP networks, future network researchers will actively study new network architectures and key technologies on the one hand to try to solve these problems. The problem is that although there are many research directions, a clear and consensus technical route has not been formed; on the other hand, when the technical route is not clear and new solutions are emerging one after another, it is necessary to establish an ultra-large-scale future network technology experiment. The verification environment (test network), on this test network, flexibly provides a resource-independent test environment for various technical solutions, thereby incubating optimal technologies. The United States, Europe and other countries have established GENI and FIRE networks respectively, and this is the purpose. During the construction of the experimental network, designers hope to be able to flexibly control and deploy routing protocols on the network nodes to achieve efficient forwarding. Therefore, there is an increasingly strong demand for the separation of the control plane and the forwarding plane on the experimental network nodes. Through the separation of the control plane, the intelligence and concentration of network control functions can be realized, as well as the protocol independence and efficiency of network forwarding functions.
In the case of separation of the control plane and the forwarding plane on the experimental network nodes, every time a new network architecture and solution appears, it can be designed and configured in the form of software on the implementation node to quickly realize a new network form. Efficiently support network technology innovation and verification.
In addition to the above three SDN requirements and the corresponding three types of experts, there are some experts who understand SDN as a unified intelligent network management, and are committed to achieving the purpose of a network management system that can manage multiple network devices in a unified and intelligent manner, such as deployment in LTE In the design of IP RAN, multiple simplified edge routers can be configured and managed through an integrated network management system, thereby improving the efficiency of network strategy deployment. However, this understanding is to separate the management plane from the control plane and the data plane, not the separation of the control plane and the forwarding plane, and should not be understood as an SDN technology.
2. "Original and Clear Source" SDN
The above analysis of the three development needs of SDN, on the whole, these needs are all pursuing the concept of network "open".
The opening of the network is an inevitable trend of industrial development. It can not only bring about the efficiency of corresponding equipment and networks, but also can further subdivide the industrial chain and bring new industrial development opportunities. At that time, the "opening" of mechanical parts realized the standard interchange between parts, refined the industrial chain of mechanical processing, improved the production efficiency of finished machinery, and greatly promoted the development of the industrial revolution. In the field of network communications, similar expectations can be achieved through SDN technology.
In the data device, it can be summarized into two planes, as shown in the following figure:
To understand SDN from the perspective of network openness, we can divide SDN into three categories, and the openness between each category is an increasing relationship:
Xiaogan Yueneng Electronic Technology Co., Ltd. , https://www.xyeloadcell.com