Shulou(Shulou.com)06/01 Report--

SDN needs and drivers

Many network colleagues will have a series of questions after they first understand the architecture of SDN(Software Defined Networking): ① Where is the real customer demand of SDN, and what problems can it solve? What is the essential difference between centralized control and centralized network management? Is it similar to soft switching? These doubts are not unreasonable. Can a simple separation of control and forwarding produce many magical effects? What I want to say is that SDN is not magical at all, it is an inevitable demand from the IT field, and it is an inevitable trend in the past 60 years that IT has become more and more hardware-oriented and software has gained functional flexibility. SDN cannot turn decadence into magic, but only adds a more flexible network component to the IT industry. In part, it seems to be a disruptive implementation technology, but in general it does not change the architecture of the existing network. It only provides a platform for enterprises and operators other than equipment suppliers to control the network to innovate on their own, reducing the cycle of network innovation from several years to several weeks. In other words, enterprises, Operators innovate to meet the needs of end users, both internal and external, and network innovation cycles are shortened and simplified, which means they are more competitive and their customers are more satisfied.

SDN has been more than 4 years since the earliest Ethane system, and it is the demand from different groups that promotes its vigorous development.

Stanford SANE/Ethane Project

Requirements: Improve network security and reduce network management complexity

Solution: Centralized control management for enterprise networks, where each endpoint must authenticate with the network, security policies can be controlled from a single point, and network devices are responsible for enforcing policies.

campus network

Demand: College researchers 'network innovations can be experimented with on campus networks without having to resort to equipment providers

Solution: The forwarding behavior of network devices is primitive and controlled by external computers. Researchers can program new network protocols and forwarding behaviors on computers based on general operating systems without modifying the forwarding surface devices themselves. OpenFlow was born.

google、Facebook、Yahoo! Large network service providers

Requirements: ① Complex network management problems in large data centers (10K switches). The new requirements for network features brought about by cloud computing technology innovation are difficult to meet in a short time. For example, it took 7 years for TRILL protocol standardization, and users hope that network features can be added as quickly as modifying software features themselves. (3) Resource scheduling required by IaaS service includes dynamic on-demand scheduling of network resources, which cannot be well met by existing technologies. Reduce network equipment costs.

Solution: ① centralized management, network equipment plug and play, automatic fault detection and traffic switching, automatic fault recovery processing. 2. The forwarding behavior of the network is controlled by an external control entity, and this external entity is an open system. Engineers at Google and Facebook can program to implement any desired network characteristics without having to go through the lengthy process of requirements, standardization, testing, and network access. (3) Pooling of network resources, network control entities provide interfaces to applications, and applications can control the forwarding behavior of the network according to needs.④ Standardization and homogenization of forwarding surface equipment, innovation and differentiation to controller software, forwarding surface cost can be greatly reduced as PC.

telecom operators

Requirements: ① Shorten the market cycle of new network functions, and hope CT innovation is as fast and diversified as the Internet. Reduce network management costs.

Solution: The network is further softwareized, generalized and centralized.

So why isn't the administrator doing all this? First of all, there is no network management that participates in the state and policy control of network operation, for example, you decide its forwarding policy online according to messages. The reliability and reasonable performance and delay guarantee of the operation system are the same as those of the operation system. In the telecommunication system, the reliability requirement is ≥99.999%. Second, the "network management" involved in the runtime process of the system is not a network management, it is some form of control plane. Third, some people say that there is no essential difference, I still think that is the network management, so well, that is the online network management server, the future network management.

SDN essence

When it comes to SDN, OpenFlow must be mentioned, but SDN is not equal to OpenFlow, just like the Internet is not equal to IP protocol, PSTN is not equal to signaling 7, IMS is not equal to SIP, WEB system is not equal to HTTP protocol. OpenFlow is only a protocol for the controller to control forwarding plane devices in SDN. The architecture of the controller itself, network topology algorithms, operating environment, programming tools, and integration technology with upper applications are all part of SDN, and are more core parts of the architecture. For example, stored program control is von. The core idea of Neumann computer system, as for what CPU instruction set you use is second, you can use the ancient ENIAC, IBM system360 instruction set, you can also use modern IBM Power, x86, MIPS, ARM instruction set, each computer system is von. An example of Neumann architecture.

Of course, for SDN, there must be specific implementation examples, not just abstract architecture, so the industry chose OpenFlow protocol as the instruction set, and built a series of operating systems, software, compilers, peripheral frameworks and implementations around it. Why did the industry choose OpenFlow instead of Forwarding and Control Element Separation (FORCES)? I understand that there are two main reasons: (1) forces are designed to separate forwarding control of devices, focusing on modeling existing functions rather than creating new network features.② The power of the promoters is different, forces initiated by Intel, but then it sold NP, no longer determined to advance, and OpenFlow proposed by Stanford University, the cradle of Silicon Valley, and promoted by GENI project through the test bed contract, which attracted Google, Facebook, Microsoft and other IT/Internet companies to participate, and finally formed a strong industrial alliance. In theory, we could extend forces to implement similar SDN functionality, or we could define other control protocols and forwarding models to implement SDN, but OpenFlow has taken the lead.

So what is the essence of SDN? As mentioned above, the essence of SDN is to allow users/applications to fully control the behavior of the network through software programming, so that the network software, and then agile. So why does the web need to be software and agile? Of course, in order to meet the needs of end customers more quickly, incidentally, reduce Capex and Opex. Some people may wonder, my device is already software programmable, SDN what is different? Of course it's different.

Step 1: Move from Device Provider Programmable to User Programmable

SDN makes the implementation of new network protocols completely programmable in the control plane by separating the control plane from the closed vendor equipment and completely controlling the forwarding plane behavior, while the control plane is an open, programmable environment based on a common operating system. The current implementation supports C++ and Python scripts, and does not rule out the future support of multiple scripting languages like Web programming. Therefore, IT/telecom operators/large enterprises with strong capabilities can implement new functions without resorting to vendors and standards organizations.

Step 2: From Device Programmable to Network Programmable

SDN is programmable not only for individual network nodes, but also for the entire network, as illustrated in the following diagram. The controller has a global topology and can compute routes between arbitrary endpoints and control forwarding paths. Similarly, it can also control the access rights of each endpoint, no matter which node you access from. For example, you can assign VLAN binding and 802.1x authentication to the controller, and the forwarding plane device is completely unaware.

Step 3: Seamless integration of networks and IT applications

As shown in the figure below, VDC service provisioning, VM migration and loading, Load Balancer and network policy migration and generation can be linked in real time through the coordination of the virtual data center manager, thereby further improving IT service response speed and service quality.

Furthermore, applications can set security policies and QoS for specific user traffic through the interfaces provided by SDN Controller, such as blocking a malicious user MAC address and reserving bandwidth for specific users/applications.

In addition, SDN-controlled networks are no longer constrained by the capabilities of standard protocols such as OSPF/ISIS/TRILL/SPB, and administrators can set up direct links between any two racks and put them into use immediately if needed, without being constrained by STP or the latest ECMP(Equivalent Multipath) capabilities.

Many people debate whether the value of OpenFlow is Open or Flow. In fact, Open has a point. Flow is completely the wrong name. Like most network devices, you can manage the flow of five tuples through ACL, but most of the traffic is L2/L3 lookup table forwarding. The granularity of traffic management depends entirely on the application requirements rather than the technology itself. OpenFlow/SDN provides look-up forwarding from fully wildcarded default routing to the fine-grained flow of tens of tuples, depending on the application scenario. If there is one device that must be managed in fine-grained flows, it is a security device, such as a firewall.

From the perspective of SDN, it is not the OpenFlow interface between the forwarding device and the controller that is Open and generates value, but the interface between the controller and the application, which is required for IT innovation.

SDN game

SDN is the power from IT to promote the transformation of the network, it is a network implementation technology, and IPv6 is completely different from IPv4, SDN does not change the forwarding plane encapsulation that the host can see, it is an existing network protocol/architecture and a future network support platform. It may be more like a high-level language + compiler that can be used to implement the application you want, rather than another new functional software.

SDN brings with it a potential that the leading vendors fear: standardized forwarding planes, and the resulting homogeneous competition, will cause network hardware profits to plummet. This risk exists, and it is a high probability, but there must be a part of the profit to be transferred to the chip and controller software, just like Intel and Microsoft in PC, although these two areas are not 100% the same. Some people may think that the controller software market may be eroded by the open source community, I personally think SDN controller is the core of the network, its reliability requirements are very high, except for a few like Google, Facebook, Microsoft, Amazon and other IT giants capable of network software development and maintenance, most enterprises can not build commercial networks based on open source software.

It is precisely because of the prospect of PC, chip manufacturers such as Broadcom and SDN software startups, network vendors see an opportunity, Cisco may see a threat, of course, Cisco's strength, embracing this shift will make the threat insignificant. I personally would love for Cisco to reject SDN outright, that would be a boon for other companies-the multi-billion dollar market will reshuffle, and we have opportunities, especially in the enterprise/data center market.

Over the past decade, we have witnessed the rise of open Google and the second rejuvenation of closed Apple. Of course, we should understand that closure is not Apple's success factor. Focusing and doing things to the extreme is its key success factor. Imitating its closed business model of software and hardware +AppStore integration is tantamount to moving against the trend. If we notice the decline of DEC and SUN at the same time, we will not imitate Apple that cannot be copied. We've also seen VMware go from being an obscure company to a giant at the top of the IT value chain, with cloud computing putting toy-like virtualization at the heart of IT. On the other hand, we have also witnessed two examples of self-reliance and self-humiliation, one is Nokia, the other is Kodak, ignoring market forces from consumers, ignoring or fighting big trends, no matter how brilliant you have been, will be abandoned by consumers.

The changes SDN brings are not too surprising, although many people compare it to the PC industry for mainframe/minicomputer, but its market space and target customers are destined to be less great than PC, of course, its entry threshold is much higher. If you want to compare it to a change that has happened in history, I prefer to compare it to the rise of virtualization technology. With server virtualization, you don't have to pay attention to where your compute/storage entity is, IT consolidation and resource utilization are possible; similarly, with SDN, you don't have to pay attention to what kind of black box is forwarding traffic, you just need to tell SDN controller what kind of network service I need.

Reference: network.51cto.com/network/content2013/SDNkepu/

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.